by RSS Xiaopeng Zhang  |  Sep 15, 2017  |  Filed in: Security Research


This is the second part of the FortiGuard Labs analysis of the new Poison Ivy variant, or PlugX, which was an integrated part of Poison Ivy’s code. In the first part of this analysis we introduced how this malware was installed onto victim’s systems, the techniques it used to perform anti-analysis, how it obtained the C&C server’s IP&Port from the PasteBin website, and how it communicated with its C&C server.

What we didn’t talk much about in that first blog was the control-commands that are used by this malware, partly because only a few of those commands were used during our analysis. However, as you may know, RAT malware usually has many control-commands so that attackers can effectively remotely control a victim’s machine.

So, after our initial analysis, we monitored the C&C servers and captured their packets. Fortunately, we were able to successfully collect enough attacks and packets so that we could obverse and document its behavior. In this analysis, I’m going to focus on the control-commands used by the C&C server as it attempts to penetrate the victim’s network by exploiting vulnerabilities.

Although the C&C servers have now been shut down, we found a way to decrypt the communication data from the captured packets in order to analyze its behavior.

As per my analysis, this variant of Poison Ivy eventually launches the MS17-010 (Eternal Blue) attack against the machines located inside the victim’s LAN. Let’s now take a look at how it performs this exploit.

Manage multiple modules

Before going on, however, we have to talk about how the decrypted modules are managed. From Part I we know that there are six modules in the svchost.exe program, which are connected by a doubly linked list. There is a module node in each of modules, as well as in svchost.exe. The module node is added into the doubly linked list when its module code is initialized. The header of the doubly linked list is in a global variable located in svchost.exe’s memory space (qword_2345D0 with base address 0x220000 in my case).  Below is a module node’s structure, along with some corrections to the one shown in the Part I of this analysis.

The first module (which was injected into svchost.exe when svchost.exe started) is executed in svchost.exe, and was the first one added into the doubly linked list. I call it the host module.

I named these module1, module2, etc. according to the order in which they are added into the doubly linked list,  The six modules are decrypted by the host module.

Figure 1 shows a view of the module node of the host (svchost.exe) in memory.

Figure 1. View of the host module node in memory

The host module node’s address is 0x334A20. The previous node’s address is 0x165068, and the next one is 0x51F280. The host module’s index is 0, and its module base address is 0x220000. Finally, the function table’s address is 0x334A60. Module index is important because it is also a part of the Control-Commands. We will talk more about this later.

Several functions in the host module are used to manage this doubly linked list. To manage the doubly linked list between these different modules, the author of the malware designed a named sharing memory (by calling API CreateFilemappingA) where the addresses of the manager functions are saved. So whenever it wants to manage the doubly linked list, it only needs to access all these functions from the sharing memory. BTW, the name of this sharing memory is created by calculating two current process IDs (by calling API GetCurrentProcessID, i.e. svchost.exe PID).

In Figure 2, you can see how the named sharing memory is created, and where the manager functions are saved in the sharing memory. The functions in [rax+8] and [rax+18] are called frequently during handling C&C commands. [rax+18] is the function that gets the module node from the doubly linked list using the module index, and sets module flag. [rax+8] is used to restore the module flag.

Figure 2. Code snippet of adding management functions into the named sharing memory

Here is the modules’ information in my test environment:

Control-Command Packet Structure

In order to easily understand the C&C packets, I will explain the packet structure here. As I explained in the first blog, the packet payload is encrypted. Through analyzing its decryption function, I was able to write a python function to decrypt the data. This is the same function that the host module used to decrypt those six modules, as well as the C&C server IP&Port from the PasteBin website, but different decryption keys are used.

Python decryption function:

The decrypted packet consists of two parts. The first 14H bytes are the header, and the data starts at offset 14H. The packet structure looks like this:

In the first blog I introduced commands “030001” and “030003”. Please refer here for more details. By the way, the malware uses big-endian byte order to save its data. The control command is a Dword value, whose high 16 bits are the module index, and the low 16 bits is a kind of code branch switch. Once the malware gets the command it retrieves the module node from the doubly linked list by matching the module index. It then calls the functions of that module to handle this command data.

Figure 3. All packets from C&C server are dispatched from here

Figure 3 shows the code snippet used for dispatching the C&C packets to the correct module for processing. After “call sub_193370” we got the decrypted C&C server packet in [rbp+arg_8]. “call sub_191C44” is used to get the management functions in rax from the named sharing memory. “call qword ptr [rax+18h]” is used to call one management function to get the module node from the doubly linked list using the module index in rcx i.e. high 16 bits of command. “call qword ptr [r8]” calls the first function of the function table to process the received packet.

From the above analysis you should now be able to clearly see the entire process of how the malware processes the C&C server’s packets.

Installing the “00000025” module

In my captured traffic, I was able to see many control commands. They include “00030001”, “00030002”, “00030003”, “00030004”, “00000003”, “00000001”, “00250000”, etc.

So let’s now take a look at what the “00000003” command is used for. Figure 4 shows the original received packet and the decrypted data.

Figure 4. “00000003” command data

From the command “00000003” details we know that this packet is going to be passed to the host module (its index is 0), and then be processed by the first function in the function table and the “0003” branch.

It gets the sub-command (“00000025”) as the module index to look for in that doubly linked list. So far, no module’s index is 0x25. It then replies to the C&C server with sub-command “00000040”. If the 0x25 module node exists, the sub-command is “00000000”.

The C&C server then sends back command “00000001” with a new module attached. Below is part data of this packet after decryption, where you can see that the sub-command is “00000025”. In code branch “0001” it decompresses the received module, then gets its code initialized, and finally adds it into the doubly linked list. This module’s index is 0x25, so I call it Module25.

It later sends command “00000001” with sub-command “00000000” to the C&C server to let it know that the 0x25 module was installed successfully. This module will be used to penetrate the victim’s network.

BTW, this module’s information in my test machine is:

Penetrating the victim’s LAN using EternalBlue

I’m sure that the C&C server sent commands to get the victim’s network configuration (my local IP, Gateway, DNS server), though I did not catch them.

Figure 5 is the screenshot of the network configuration of my test machine.

Figure 5. Network information

The C&C server controls the malware to scan the victim’s network segment, including local IP, Gateway, and DNS server.  For example, because my DNS server is it’s going to scan the network segment.

The C&C server sends the “00000025” command with the destination IP and Port for further attack. By decrypting “00000025” packets we are able to see its data, shown below.

From this data it is easy to see that there are IP addresses from three local machines. The sub-command “000001BD” refers to port 445.

Module25 processes this packet, pulls the IP and port information from the packet, and then makes a connection to it. If any error occurs, it sends the status to the C&C server.

Once successfully connected to the destination machine, the malware then serves as a middleman that keeps transferring the two sockets’ data between the C&C server and the destination machine (like man-in-the-middle does). In module3 we also see its debug output strings “SoTransfer(%p<=>%p)...\r\n” and “SoTransfer(%p<=>%p) quit!\r\n”. Figure 6 and 7 show the attack view in Wireshark.

Figure 6. EternalBlue attack packets

Figure 7. EternalBlue attack packet payload

Module25 makes the connection to the destination IP and then calls module3’s function to perform the transfer work by calling the recv() and send() functions. In module3 function sub_1935A8  it creates two threads to do that. One thread receives data from the C&C socket and sends it to the destination machine, and another one receives data from the destination machine and forwards it to the C&C server.  Figure 8 shows the code snippet for what I explained about the two threads.

Figure 8. Two threads to transfer packets


Based on our analysis, this new Poison Ivy variant takes advantage of the EternalBlue exploit to spread. Once one system is infected by this variant, other systems on the same network are likely to be infected by the compromised system.


Users should apply Microsoft’s patch for MS17-010.

Fortinet IPS signature MS.SMB.Server.SMB1.Trans2.Secondary.Handling.Code.Execution was released in March 2017 to protect our customers against the EternalBlue attack.

Sign up for weekly Fortinet FortiGuard Labs Threat Intelligence Briefs and stay on top of the newest emerging threats.

by RSS Xiaopeng Zhang  |  Sep 15, 2017  |  Filed in: Security Research