NetworkMiner 2.9 Released

NetworkMiner 2.9 brings several new and improved features to help analysts make sense of network traffic from malware, criminals and industrial control systems. Highlights from this new version include:

• TZSP support
• StealC extractor
• Improved Modbus parser
• JA4 support
• GTP decapsulation

Malware Traffic Artifact Extraction

NetworkMiner is a popular tool for extracting artifacts from malware traffic. Such artifacts can be downloaded malware modules, exfiltrated documents and sometimes even screenshots of the infected computer.

Parsers for njRAT and BackConnect (à la IcedID, QakBot and Bazar) traffic was previously added to NetworkMiner. In this release NetworkMiner also gets a parser for StealC, which has quickly become one of the most popular information stealers on Russian-speaking underground forums. The new NetworkMiner 2.9 release extracts screenshots and files that SteakC exfiltrates from the infected machine.

The examples shown below were created by loading a pcap file with StealC traffic from Triage sandbox into NetworkMiner 2.9. NetworkMiner was run in Linux to minimize the risk of accidentally infecting the analysis environment.

Image: Reassembled system info and documents exfiltrated by StealC to 185.172.128.151

Image: Reassembled screenshot of victim’s desktop sent to StealC C2 server

NetworkMiner’s VNC and BackConnect VNC parser has also been improved in this release. NetworkMiner’s keylog extraction from VNC now supports lots of keyboard layouts, including Arabic, Cyrillic, Greek, Hebrew, Kana, Korean and Thai. The handling of VNC color profiles has also been improved to convey colors more correctly in screenshots from reassembled VNC and BackConnect VNC traffic. I’d like to thank Brad Duncan and Maxime Thiebaut for their valuable input on this matter!

Another remote management tool that often is misused by hackers and criminals is Remote Manipulator System (RMS) from TektonIT. According to Cyberint’s report Legit remote admin tools turn into threat actors’ tools there are lots of Russian forum posts and even YouTube tutorials showing how to include legitimate RMS components in malware. NetworkMiner now parses RMS’s session setup, which includes information about the client computer as well as the RMS product and version. The screenshot below was created by loading a pcap file from when 3_Рахунок.pdf.exe was executed in JoeSandbox.

Image: Information extracted from RMS traffic

The country_code number (here 223) also gets converted to a human-readable country (Switzerland) by NetworkMiner, but this country name info is only displayed in the Host Details of the client.

ICS / SCADA

NetworkMiner has supported Modbus/TCP since 2016 (when NetworkMiner 2.0 was released). This Modbus parser has now been updated to display Modbus addresses using the Modicon convention, which explicitly specifies the register type while also signalling to the user that the displayed addresses are one-indexed.

The register types are displayed in parenthesis and should be interpreted as follows:

• (0)nnnn = Coil
• (1)nnnn = Discrete input
• (3)nnnn = Input register
• (4)nnnn = Holding register

NetworkMiner now also reads Modbus Device Identification messages and displays the reported device info in Host Details. This feature is very handy if you’re building an asset inventory through passive asset discovery (i.e. passively monitoring traffic to identify devices).

NetworkMiner 2.9 also supports asset identification for ICS networks that use COTP based protocols, such as Siemens S7 protocol or IEC 61850 MMS, by parsing COTP connection request messages. The identified COTP TSAP names are displayed under Host Details.

Image: NetworkMiner showing a WinCC client and a Siemens SIMATIC device

User Interface Improvements

TLS handshake fingerprinting with JA3 was added to NetworkMiner in 2019, but last year John Althouse announced the new JA4+ fingerprint methods. In short JA4+ is a suite of methods designed to fingerprint implementations of a specific set of protocols, including TLS, HTTP and SSH. Most of the fingerprinting methods in the JA4+ suite are patent pending except for the TLS client fingerprinting method JA4, which is an improved version of JA3. NetworkMiner now generates both JA3 and JA4 fingerprints for TLS handshakes. The results from the TLS fingerprinting can be seen in the Parameters tab as well as Host Details. In the example below we’ve loaded TLS traffic to port 8533 on 91.92.251.26 from a Remcos sample on ANY.RUN into NetworkMiner Professional (the free NetworkMiner edition doesn’t parse TLS traffic to non-standard ports).

Image: JA4 hash t13i010400_0f2cb44170f4_5c4c70b73fa0 for Remcos C2 traffic

NetworkMiner has also been improved to extract even more information from HTTP traffic, such as JSON formatted parameters and telemetry data sent to Microsoft by their Device Metadata Retrieval Client (DMRC). We have also improved the DNS extraction, both with regards to DNS TXT labels and Additional Resource Records.

The previous Remcos screenshot displays a latency measurement (0.0935 ms), which is another new feature in this release. This latency value is an estimation of the average timespan from when the host sends a packet until it gets captured by the sniffer. NetworkMiner’s hosts list can be sorted based on the Latency value, whereby local computers and network devices are shown at the top of the list. Another way to achieve similar results is to instead sort the hosts based on “Router Hops Distance”.

NetworkMiner’s user interface has also been improved to make it easier to copy text from the Hosts and Parameters tabs with Ctrl+C or by right-clicking and selecting “Copy …”. The export-to-file function in NetworkMiner Professional now additionally includes data from the Keywords tab.

TZSP Sniffing and Decapsulation

Routers from Mikrotik have a feature called TZSP (short for TaZmen sniffer Protocol), which encapsulates captured traffic into TZSP packets and then transmits them to a streaming server. This feature is similar to PCAP-over-IP and ERSPAN, except TZSP transports the sniffed packets over UDP instead of TCP or GRE.

NetworkMiner now includes a TZSP streaming server, which can receive TZSP encapsulated traffic over a UDP socket. Click “File, Receive TZSP Stream”, select a port (default is 37008) and click “Start” to receive a real-time stream of captured packets from a Mikrotik router. We’ve also added support for TZSP link layer type (DLT_TZSP) pcap files as well as decapsulation of TZSP packets to UDP port 37008. I’d like to thank Jarmo Lahtiranta for proposing this feature!

Speaking of decapsulation – we’ve added a GTP-U parser, which enables NetworkMiner to analyze GPRS traffic from GSM, UMTS, LTE and 5G networks that is transmitted inside a GTP tunnel.

Upgrading to Version 2.9

Users who have purchased NetworkMiner Professional can download version 2.9 from our customer portal, or use the “Check for Updates” feature from NetworkMiner's Help menu. Those who instead prefer to use the free and open source version can grab the latest version of NetworkMiner from the official NetworkMiner page.

Posted by Erik Hjelmvik on Monday, 27 May 2024 09:50:00 (UTC/GMT)

Tags: #NetworkMiner​ #TZSP​ #Modbus​ #JA4​ #BackConnect​ #VNC​ #JSON​

Forensic Timeline of an IcedID Infection

The BackConnect and VNC parsers that were added to NetworkMiner 2.8.1 provide a unique possibility to trace the steps of an attacker with help of captured network traffic from a hacked computer.

In this blog post I use the free and open source version of NetworkMiner to see how GzipLoader downloads IcedID, after which the attacker deploys BackConnect VNC to purchase an iPhone 14 with a stolen credit card and then drops Cobalt Strike on the victim PC.

The analyzed pcap is 2022-10-31-IcedID-with-DarkVNC-and-Cobalt-Strike-full-pcap-raw.pcap from Brad Duncan's malware-traffic-analysis.net blog.

Safety First

I ran NetworkMiner in a Windows Sandbox when analyzing this PCAP file to avoid accidentally infecting my computer with any of the malicious artifacts that NetworkMiner extracts from the network traffic.

Another safe way to analyze Windows malware is to run NetworkMiner in Linux or macOS.

14:47 GzipLoader

This infection starts with GzipLoader (aka “IcedID Downloader”) reaching out to its C2 server on vgiragdoffy[.]com (67.205.184.237:80) to download IcedID.

The “_gat” cookie value in frame number 6 tells us that the victim machine is running a Windows 10 build 19045 (aka 22H2). The long “_u” value contains the victim’s username and hostname in hexadecimal representation and the “__io” value is the logged in user’s SID. NetworkMiner decodes these values from the GzipLoader request and displays them in the Hosts tab.

For more info about the GzipLoader cookie, see IcedID PhotoLoader evolution by Jason Reaves and the eSentire blog post on Gootloader and IcedID.

The response for this GzipLoader request is a 550 kB file (MD5 700c602086590b05dde8df57933c7e68) with a fake gzip header. This file actually contains the IcedID DLL (Odwikp.dll) and license.dat files.

14:47 IcedID

The banking trojan IcedID (aka BokBot) gets launched at 14:47:29 UTC (frame 641) after which it connects to these four IcedID servers used for payload delivery and C2:

• ringashopsu[.]com = 137.184.208.116
• sainforgromset[.]com = 138.68.255.102
• yeloypod[.]hair = 94.140.114.103
• airsaintol[.]beauty = 66.63.168.75

These four IcedID servers all run TLS servers with self signed certificates issued for "localhost" and doing TLS handshakes with JA3S hash ec74a5c51106f0419184d0dd08fb05bc. Both these properties can be used as filters in NetworkMiner's Hosts tab to only display the IcedID C2 servers.

14:59 BackConnect and Keyhole VNC

Shortly after the IcedID C2 traffic has been started the IcedID bot also initiates BackConnect C2 connections to 137.74.104.108 on TCP port 8080 (frame 4505 at 14:59:14 UTC).

The BackConnect C2 server tells the bot to sleep for 60 seconds two times before launching a reverse VNC session with command 0x11 (frame 4530 at 15:01.09 UTC).

15:06 Apple Store

The keylog of the attacker above reveals that the attacker is typing “iphone 14 apple store buy”. The VNC graphics that NetworkMiner extracted from the PCAP file additionally reveal that this was a Google search query typed into an Edge browser.

15:10 Credit Card payment

The attacker proceeds to the Apple Store, puts a black iPhone 14 Plus for $987.99 into the shopping cart, enters a delivery address in West Hartford (US) and then inputs credit card details for the payment.

Luckily, the transaction was denied by Apple Store.

15:12 Reverse Shell

After having failed to buy an iPhone through the hacked computer the attacker instead deploys three reverse shell sessions using the BackConnect C2 channel.

These three commands are issued in the first reverse shell session:

In the second shell session the attacker first runs these three commands:

...and then starts a file manager session through the BackConnect C2 channel.

15:40 Deploy Cobalt Strike

The BackConnect file manager is used to upload a Cobalt Strike binary called P2.dll to "C:\ProgramData\" on the victim computer in frame 144535.

The uploaded P2.dll is then executed by running this command in the reverse shell session (frame 144707):

NetworkMiner extracts this uploaded DLL from the BackConnect network traffic.

As you can see in the screenshot above, the MD5 hash of P2.dll is cc69a31a067b62dda5f2076f8ee335e1. This file is flagged as malicious by most AV vendors (P2.dll on VT). However, none of them label it as Cobalt Strike. Luckily I was able to use Triage's malware config extractor to verify that this was indeed Cobalt Strike (P2.dll on tria.ge). Triage also revealed that the CobaltStrike C2 URL was
clouditsoft[.]com:8008/static-directory/mg.jpg

After the DLL gets executed the victim PC establishes Cobalt Strike beacon C2 connections to clouditsoft[.]com on port 8008 (frame 144715).

15:41 MOAR COBALT STRIKE

The BackConnect Reverse Shell log in NetworkMiner's Parameters tab shows that the attacker also attempted to download Cobalt Strike using PowerShell at 15:41:59 UTC (frame 145176) with this command:

IOC List

• IP:port 67.205.184.237:80 (GzipLoader)
• DNS vgiragdoffy[.]com (GzipLoader)
• MD5 700c602086590b05dde8df57933c7e68 (Fake gzip file)
• MD5 f57ab2e5e5720572d5eb19010ec8dcb4 (IcedID Odwikp.dll from fake gzip)
• MD5 57a9d9acb389bd74a7423a16ef81ac18 (IcedID license.dat from fake gzip)
• DNS ringashopsu[.]com (IcedID C2)
• DNS sainforgromset[.]com (IcedID C2)
• DNS yeloypod[.]hair (IcedID C2)
• DNS airsaintol[.]beauty (IcedID C2)
• IP:port 137.184.208.116:443 (IcedID C2)
• IP:port 138.68.255.102:443(IcedID C2)
• IP:port 94.140.114.103:443 (IcedID C2)
• IP:port 66.63.168.7:443 (IcedID C2)
• JA3S hash ec74a5c51106f0419184d0dd08fb05bc (IcedID C2)
• IP:port 137.74.104.108:8080 (IcedID BackConnect C2)
• MD5 cc69a31a067b62dda5f2076f8ee335e1 (CobaltStrike P2.dll)
• DNS clouditsoft[.]com (CobaltStrike C2)
• IP:port 198.44.140.67:8008 (CobaltStrike C2)

Posted by Erik Hjelmvik on Thursday, 12 October 2023 13:23:00 (UTC/GMT)

Tags: #NetworkMiner​ #IcedID​ #GzipLoader​ #BackConnect​ #VNC​ #CobaltStrike​ #Cobalt Strike​ #Windows Sandbox​ #ec74a5c51106f0419184d0dd08fb05bc​ #JA3S​

NetworkMiner 2.8.1 Released

I am happy to announce the release of NetworkMiner 2.8.1 today!

This new release brings a VNC parser to NetworkMiner, so that screenshots, keystrokes and clipboard data can be extracted from unencrypted VNC traffic. NetworkMiner 2.8.1 additionally includes parsers for command-and-control (C2) protocols used by njRAT, IcedID, QakBot and Bazar. We have also added extraction of files sent using a HTTP PUT request, which is the rare cousin to the more common POST request. NetworkMiner’s Images tab has also been updated to enable filtering based on filename or pixel count.

VNC

NetworkMiner now supports the remote framebuffer (RFB) protocol (RFC 6143), which is the underlying protocol used in VNC software. VNC is used to remotely control another computer with help of a real-time video feed of the remote computer’s screen and the ability to control the remote desktop using a mouse and keyboard.

NetworkMiner reassembles the desktop graphics that get transferred between a VNC client and server to build a local representation of the remote desktop. Screenshots of that desktop then get extracted to disk by NetworkMiner as it gets updated. These desktop screenshots can be viewed in NetworkMiner’s Images tab. NetworkMiner also extracts keystrokes and clipboard data from the VNC traffic and displays them in the Parameters tab.

Limitations in VNC Extraction

Essential configuration of the VNC session is set up in the beginning of each TCP session. NetworkMiner therefore requires analyzed traffic to include this initial setup in order to successfully extract graphics from a VNC session.

Only traffic to TCP ports 5900 and 5901 will be parsed as VNC traffic by the free version of NetworkMiner. Our commercial tool NetworkMiner Professional has the ability to identify VNC traffic regardless of port number, but the free version will fail to extract screenshots from VNC traffic running on non-standard ports.

NetworkMiner’s VNC implementation is currently at an early stage, which is why it might fail to extract screenshots or other data from VNC sessions. If you encounter VNC traffic that NetworkMiner fails to parse, then we’d be very grateful if you would contact us and let us know so that we can improve NetworkMiner’s VNC extraction in future releases.

BackConnect

I have previously outlined the BackConnect protocol used by IcedID, QakBot as well as Bazar. This BackConnect protocol is used by botnet operators to remotely control a victim computer using a form of remote access trojan (RAT). This BackConnect protocol supports features such as tunneling network traffic through the victim’s computer, uploading files and the ability to control the victim’s computer using VNC.

NetworkMiner now has a parser for this BackConnect protocol, which allows analysts to see screenshots, keystrokes and file uploads from when the attacker interacts with the hacked computer. As you might have guessed, NetworkMiner’s VNC parser is reused in order to also parse reverse VNC traffic from BackConnect sessions.

I would like to thank Maxime Thiebaut for releasing his PCAPeek tool, which also reassembles reverse VNC traffic from BackConnect traffic. Maxime’s tool proved that it was possible to extract graphics from this C2 traffic, which motivated me to create a BackConnect VNC parser for NetworkMiner.

BackConnect VNC Replay

To demonstrate NetworkMiner’s reverse VNC reassembly I’d like to use the file “2023-07-26-IcedID-with-Keyhole-VNC-traffic.pcap” from Brad Duncan’s fantastic malware-traffic-analysis.net website. In this traffic the IcedID BackConnect C2 server listens to TCP port 443 on 165.232.175.216.

The screenshot thumbnails from the BackConnect VNC traffic shows that the attacker opens up the Amazon website on the victim’s computer. A copy of the full 1920x1152 screenshot, which NetworkMiner extracted from the VNC traffic, can be viewed here: VNC_73F81C83_230726142247.jpg

The parameters tab in NetworkMiner additionally confirms that the attacker manually typed “amazon.com” into the hacked computer’s web browser.

Limitations in BackConnect Parser

The BackConnect implementation in NetworkMiner 2.8.1 is pretty well tested and even supports the new “encrypted” version of BackConnect that IcedID has been deploying lately (we call it “BackConnect XOR”). NetworkMiner can parse the BackConnect C2 traffic as well as the traffic from modules for reverse VNC, reverse SOCKS, reverse shell and the file manager.

Unfortunately the BackConnect traffic typically runs on TCP ports like 443 or 8080, which are normally used for TLS or web proxies. The free open source version of NetworkMiner normally attempts to interpret traffic to port 443 as TLS and 8080 as HTTP proxy, even though it might be BackConnect. A hardcoded list of 32 verified BackConnect C2 servers, such as 165.232.175.216:443, 137.184.172.23:443 and 185.99.132.16:8080, has therefore been added to NetworkMiner’s source code. This enables NetworkMiner to interpret traffic to those IP:port combinations as BackConnect. Our commercial tool NetworkMiner Professional can identify protocols without relying on the port number though, this enables the “Pro” version to also parse BackConnect traffic to C2 servers that aren’t among the 32 hard coded ones in the source code.

njRAT

njRAT is an old, yet still very popular remote access trojan (RAT). It has been around for at least 10 years (since 2013) and can be used to do things like:

• Steal credentials
• Take screenshots of the desktop
• Start the camera
• Record audio with the microphone
• Start a reverse shell
• Upload or download files
• Execute a file or module
• Capture keystrokes with a keylogger
• Create, read or delete registry keys
• Run a command

I have been teaching students in my network forensics classes how to manually parse njRAT traffic since 2015, but from now on I’ll instead be able to tell them to load the njRAT traffic into NetworkMiner to see what the attacker did on the victim’s computer.

The NetworkMiner parameters and victim PC screenshot above were extracted from 2015-04-22_capture-win4.pcap in StratosphereIPS’ dataset CTU-Malware-Capture-Botnet-120-1. The victim PC desktop screenshot actually came out as a 51x4900 pixel image in NetworkMiner 2.8.1, so I had to manually reorganize the tiles to a 510x490 image instead. I’m planning to add an automatic retiling feature to NetworkMiner to avoid having to do this manually in the future.

Limitations in njRAT Parser

njRAT comes in many different versions, forks and flavors and the protocol even has a field delimiter sequence that can be customized for each njRAT backdoor. I can therefore not guarantee that NetworkMiner will be able to parse traffic from all njRAT sessions, but so far I’ve seen very good results – even when custom field delimiters are being used.

Nevertheless, this trojan has an abundance of features but only the most commonly used ones are implemented in NetworkMiner. More specifically NetworkMiner primarily supports extraction of screenshots, credentials, keylog data, file uploads and downloads. NetworkMiner also extracts exfiltrated metadata about the victim machine and bot, such as the hostname, operating system, logged in user, bot ID, botnet, njRAT version and njRAT install date.

Also, as previously mentioned, NetworkMiner doesn’t yet stitch together tiles of desktop screenshots correctly.

Another limitation is that an njRAT C2 server can be deployed on any TCP port, which might prevent the free version from interpreting the C2 traffic as njRAT. The free version of NetworkMiner currently tries to parse traffic to ports like 1177, 5050, 5552 and a few others as njRAT, while NetworkMiner Professional will automatically recognize the njRAT protocol regardless which port it runs on.

Bug Fixes

We’ve added support for SIP keepalive messages (RFC 5626) in NetworkMiner 2.8.1. NetworkMiner Professional would previously fail to extract VoIP calls, or text messages sent through SIP, if a keepalive packet was sent in the session prior to the call or text message. We have also fixed a bug in NetworkMiner’s CapWap parser.

Upgrading to Version 2.8.1

Users who have purchased NetworkMiner Professional can download a free update to version 2.8.1 from our customer portal, or use the “Check for Updates” feature from NetworkMiner's Help menu. Those who instead prefer to use the free and open source version can grab the latest version of NetworkMiner from the official NetworkMiner page.

Posted by Erik Hjelmvik on Monday, 02 October 2023 11:13:00 (UTC/GMT)

Tags: #NetworkMiner​ #VNC​ #BackConnect​ #IcedID​ #njRAT​

IcedID BackConnect Protocol

This is a follow-up to my Hunting for C2 Traffic video. But I didn't have time to record a short video this time, so I wrote a long blog post instead.

UPDATE 2022-11-02

Brad Duncan has released a new pcap file on malware-traffic-analysis.net, which contains an additional C2 command (0x12). Our analysis indicates that this command launches a file manager. This blog post has now been updated with details about this finding.

UPDATE 2022-11-09

Lenny Hansson has released IDS signatures that detect BackConnect traffic. More details further down in this blog post.

UPDATE 2022-12-05

Lenny has updated his IDS signatures to alert on BackConnect C2 traffic from port 443 in addition to 8080. The signatures in this blog post have now been updated to Lenny's new rev:2 signatures.

UPDATE 2023-04-14

Brad Duncan made the following suggestion in a toot yesterday:

If the protocol for this VNC traffic from Qakbot looks the same as the BackConnect traffic from IcedID infections, perhaps we can just call it BackConnect Protocol without specifying "IcedID"

This is an excellent idea, since IcedID, QakBot as well as Bazar have all been seen using the same BackConnect protocol. We will therefore refer to the protocol described in this blog post as just the “BackConnect Protocol” from now on. This blog post has also been updated accordingly.

UPDATE 2023-10-02

The release of NetworkMiner 2.8.1 adds a BackConnect protocol parser to NetworkMiner.

IcedID BackConnect C2 Packet Structure

The BackConnect (BC) module uses a proprietary command-and-control (C2) protocol that is pretty straight forward. Both client (bot) and the C2 server typically send commands and responses as 13 byte packets using the following structure:

• Auth: 4 bytes
• Command: 1 byte
• Params: 4 bytes
• ID: 4 bytes

Auth Field

The "Auth" field is presumably used by the bot and C2 server to verify that the other party is communicating using the same protocol and version.

As mentioned by Group-IB and xors the Auth field is typically 0x974F014A (little endian), but we prefer to use the network byte order representation "4a 01 4f 97".

In their IcedID blog post from 2020 Group-IB say:

the auth field that has not changed since at least version 5 of the IcedID core is the constant 0x974F014A

Nevertheless, we recently noticed another BackConnect Auth field being used in the wild. But more on that later.

Commands

The following list of BackConnect C2 commands has been compiled by combining those mentioned by Group-IB with our own analysis of the BackConnect protocol:

• 0x00 = Bot queries for a task
• 0x01 = Set sleep timer
• 0x02 = Bot error
• 0x03 = Reconnect
• 0x04 = Start SOCKS
• 0x05 = Start VNC

We've also discovered these additional commands in BackConnect C2 traffic that uses the Auth value "1f 8b 08 08":

• 0x11 = Start VNC
• 0x12 = Start file manager
• 0x13 = Start reverse shell

Commands 0x04, 0x05, 0x11, 0x12 and 0x13 all cause the bot to connect back to the C2 server using a new BackConnect session, which will be used to wrap either SOCKS, VNC, file manager or reverse shell traffic.

Command 0x01: Set Sleep Timer

The set sleep timer command is issued by the C2 server to instruct the bot to sleep for a certain amount of time before requesting a new task from the C2 server again. The sleep time is defined in the four bytes following directly after the 0x01 command. This value is a 32-bit little endian value indicating the number of seconds the bot should sleep, i.e. "3c 00 00 00" = 0x0000003c = 60 seconds. The most common sleep value seems to be 60 seconds, which is why you'll often see byte sequences like this in IcedID C2 sessions:

zz zz zz zz 01 3c 00 00 00 xx xx xx xx

The following Wireshark display filter will show BackConnect C2 packets, where the bot is configured to sleep for 60 seconds before querying the C2 server for a new command:

tcp.len == 13 and tcp.payload[4:5] == 01:3c:00:00:00

Command 0x04: Start SOCKS

The SOCKS command (0x04) instructs the bot to start the SOCKS module. As an example, the following byte sequence was sent by the IcedID C2 server 91.238.50.80:8080 in Brad Duncan's 2022-06-28 TA578 IcedID pcap on malware-traffic-analysis.net (see frame #10231):

4a 01 4f 97 04 09 00 00 00 8c a2 b1 09

The first four bytes are the auth value, followed by the Start SOCKS command (04).

After receiving this command the bot established a new TCP connection back to the C2 server, where it echoed back the server's "Start SOCKS" command and then started acting like a SOCKS server.

Except for initially echoing the BackConnect Start SOCKS command the SOCKS module actually seems to be compliant with RFC1928, which defines the SOCKS5 protocol. This means that the C2 server can supply an IP address and port number to the bot's SOCKS proxy in order to relay a connection to that host through the bot.

Image: C2 server instructs bot to relay a connection to 188.40.30.100:80

After receiving a Start SOCKS command an IcedID bot immediately establishes a new TCP connection to the specified IP and port, and relays the application layer data back to the C2 server through the SOCKS connection.

Image: Update check of Advanced Port Scanner relayed through the infected machine

In the 2022-06-28 TA578 IcedID pcap the attacker used multiple SOCKS connections to scan the 10.6.21.0/24 network for services running on TCP ports 21, 80, 445 and 4899. That last port (TCP 4899) is typically used by Radmin VPN, which just so happens to be created by the outfit "Famatech" who also develop the "Advanced Port Scanner". The attacker also used the SOCKS module to make several HTTPS connections to servers like 18.204.62.252 (tlx.3lift[.]com), 23.94.138.115 (cmd5[.]org) and 74.119.118.137 (cat.da.us.criteo[.]com). The attacker also proxied connections to 40.97.120.242 and 52.96.182.162 (outlook.live.com) through the infected bot.

NetworkMiner showing hosts that the bot proxied TLS traffic to

JA3 Fingerprints from Proxied Traffic

Since the SOCKS proxy doesn't touch the application layer data we know that the client TLS handshake packets are coming from the C2 server rather than from the bot that's running the SOCKS proxy. This means that we can fingerprint the actual TLS client using JA3.

As you can see in the CapLoader screenshot above, most proxied TLS sessions use the cd08e31494f9531f560d64c695473da9 JA3 hash, but two of them use the rare JA3 hash 598872011444709307b861ae817a4b60. That rare JA3 hash was used only when connecting to outlook.live.com.

Command 0x05 or 0x11: VNC

Brad Duncan's 2022-06-28 TA578 IcedID pcap also contains the "Start VNC" command 0x05.

Image: Flow transcript of Start VNC command

As can be seen in the CapLoader screenshot above, Start VNC commands were sent at 16:33:33 and 16:34:06 UTC. And just like the SOCKS command, this caused the bot to establish a new connection back to the C2 server, echo the "Start VNC" command and then proceed with the VNC traffic.

Image: Flow transcript of IcedID VNC traffic in ASCII encoding

Command 0x13: Reverse Shell

Brad posted a new capture file with network traffic from another IcedID infection last week (2022-10-04). He also noted that the traffic to 51.89.201.236:8080 was different from normal IcedID post-infection traffic.

After looking at this C2 traffic I discovered that it was in fact using the IcedID BackConnect protocol outlined in this blog post, but the Auth field "4a 01 4f 97" had been replaced with "1f 8b 08 08".

That exact byte sequence is a common file header for gzip compressed files (RFC1952), where

• 1f 8b = GZIP magic
• 08 = DEFLATE compression
• 08 = Original file name header present

IcedID has previously been seen using fake gzip file headers in payloads, but this time even the C2 packets include the gzip header!

Image: Transcript of TCP session to 51.89.201.236:8080

The C2 traffic also contained the command 0x13, which I hadn't seen before. Just like the SOCKS and VNC commands, this one triggered the bot to establish a new connection back to the C2 server. But the bot sent a task query command (00) this time, instead of echoing the C2 server's command (0x13). The new TCP session then transitioned into what looks like a reverse shell session.

Image: Transcript of reverse shell traffic from IcedID BackConnect session

The reverse shell traffic reveals that the attackers retrieved a list of domain admin users and then executed a PowerShell script from aicsoftware[.]com. This PowerShell script was used to install CobaltStrike beacon on the victim's PC.

Command 0x12: File Manager

We discovered the file manager command after this blog post was published. This section has therefore been added after the original publication of this blog post.

The following Wireshark display filter can be used to find file manager commands (0x12) in BackConnect C2 traffic that uses the "1f 8b 08 08" auth value:

tcp.len == 13 and tcp.payload[0:5] == 1f:8b:08:08:12

Image: File manager commands in BackConnect C2

The screenshot above shows that the file manager command was issued three times in 2022-10-31-IcedID-with-DarkVNC-and-Cobalt-Strike-full-pcap-raw.pcap.

Image: BackConnect TCP sessions in CapLoader's Flows view

As you can see in the two screenshots above, each time a file manager command was issued in the C2 session (Wireshark screenshot) the bot established a new TCP connection back to the C2 server (CapLoader screenshot).

The file manager sessions use a proprietary protocol to perform tasks such as listing disks, changing directory and uploading files.

We've identified the following file manager commands:

• DISK = List drives
• CDDIR <path> = Change directory
• PWD = Show current directory
• DIR = List current directory
• PUT <path> = Upload file

IDS Signatures

Lenny Hansson has released IDS signatures that can detect IcedID (and QakBot) BackConnect traffic. I'd like to highlight four of Lenny's signatures here.

Alert on "sleep 60 seconds" C2 command, regardless of Auth value:

alert tcp $EXTERNAL_NET [443,8080] -> $HOME_NET 1024: (msg:"NF - Malware IcedID BackConnect - Wait Command"; flow:established; flags:AP; dsize:13; content:"|01 3c 00 00 00|"; offset:4; depth:5; reference:url,networkforensic.dk; metadata:02112022; classtype:trojan-activity; sid:5006006; rev:3;)

Alert on "start VNC" C2 command with "4a 01 4f 97" Auth:

alert tcp $EXTERNAL_NET [443,8080] -> $HOME_NET 1024: (msg:"NF - Malware IcedID BackConnect - Start VNC command"; flow:established; flags:AP; dsize:13; content:"|4a 01 4f 97 05|"; offset:0; depth:5; reference:url,networkforensic.dk; metadata:03112022; classtype:trojan-activity; sid:5006007; rev:2;)

Alert on "start VNC" C2 command with "1f 8b 08 08" Auth:

alert tcp $EXTERNAL_NET [443,8080] -> $HOME_NET 1024: (msg:"NF - Malware IcedID BackConnect - Start VNC command - 11"; flow:established; flags:AP; dsize:13; content:"|1f 8b 08 08 11|"; offset:0; depth:5; reference:url,networkforensic.dk; metadata:03112022; classtype:trojan-activity; sid:5006011; rev:2;)

Alert on "start file manager" C2 command with "1f 8b 08 08" Auth:

alert tcp $EXTERNAL_NET [443,8080] -> $HOME_NET 1024: (msg:"NF - Malware IcedID BackConnect - Start file manager command"; flow:established; flags:AP; dsize:13; content:"|1f 8b 08 08 12|"; offset:0; depth:5; reference:url,networkforensic.dk; metadata:03112022; classtype:trojan-activity; sid:5006008; rev:2;)

A zip file containing Lenny's Snort rules can be downloaded from networkforensic.dk.

Questions and Answers

Allright, that's all I had to say about the IcedID BackConnect C2 protocol. I'm now ready to take your questions.

Q: Is IcedID's BackConnect VNC traffic the same thing as DarkVNC?

No, DarkVNC traffic doesn't use the BackConnect C2 Packet Structure described in this blog post. Also, one characteristic behavior DarkVNC is that the first C2 packet contains a string that looks like one of these:

• (COMPUTERNAME)_ADDITIONAL_ID-DARKVNC
• BOT-COMPUTERNAME(USERNAME)_ID-REFnnn
• USR-COMPUTERNAME(USERNAME)_ID-REFnnn

Q: Is IcedID's BackConnect VNC traffic the same thing as hVNC?

Almost. hVNC means "hidden VNC" and includes any type of malicious VNC server running on a victim's PC, including IcedID's VNC module as well as DarkVNC.

Q: How did you get Wireshark to decode the SOCKS traffic from IcedID BackConnect?

1. Open the pcap file from 2022-06-28 TA578 IcedID
2. Apply display filter: tcp.port eq 8080
3. Right-click, Decode As, TCP port 8080 = SOCKS
4. Display filter: tcp.dstport eq 8080 and tcp.len eq 13 and tcp.payload[0:5] eq 4a:01:4f:97:04
5. Select all packets (Ctrl+A)
6. Edit, Ignore Packets (Ctrl+D)
7. Display filter: socks.dst

Q: Can CapLoader's Protocol Identification feature detect the BackConnect protocol?

The version used in this blog post (1.9.4) doesn't have a protocol model for the BackConnect protocol, but later versions can identify IcedID's BackConnect protocol regardless of port. CapLoader version 1.9.5 (and later) also alerts on BackConnect traffic.

Posted by Erik Hjelmvik on Wednesday, 12 October 2022 18:24:00 (UTC/GMT)

Tags: #IcedID​ #QakBot​ #QBot​ #TA578​ #BackConnect​ #SOCKS​ #SOCKS5​ #VNC​ #JA3​ #gzip​ #PowerShell​