At the beginning of October 2017, we discovered new Android spyware with several features previously unseen in the wild. In the course of further research, we found a number of related samples that point to a long-term development process. We believe the initial versions of this malware were created at least three years ago – at the end of 2014. Since then, the implant’s functionality has been improving and remarkable new features implemented, such as the ability to record audio surroundings via the microphone when an infected device is in a specified location; the stealing of WhatsApp messages via Accessibility Services; and the ability to connect an infected device to Wi-Fi networks controlled by cybercriminals.
We observed many web landing pages that mimic the sites of mobile operators and which are used to spread the Android implants. These domains have been registered by the attackers since 2015. According to our telemetry, that was the year the distribution campaign was at its most active. The activities continue: the most recently observed domain was registered on October 31, 2017. Based on our KSN statistics, there are several infected individuals, exclusively in Italy.
Moreover, as we dived deeper into the investigation, we discovered several spyware tools for Windows that form an implant for exfiltrating sensitive data on a targeted machine. The version we found was built at the beginning of 2017, and at the moment we are not sure whether this implant has been used in the wild.
We named the malware Skygofree, because we found the word in one of the domains*.
According to the observed samples and their signatures, early versions of this Android malware were developed by the end of 2014 and the campaign has remained active ever since.
Signature of one of the earliest versions
The code and functionality have changed numerous times; from simple unobfuscated malware at the beginning to sophisticated multi-stage spyware that gives attackers full remote control of the infected device. We have examined all the detected versions, including the latest one that is signed by a certificate valid from September 14, 2017.
The implant provides the ability to grab a lot of exfiltrated data, like call records, text messages, geolocation, surrounding audio, calendar events, and other memory information stored on the device.
After manual launch, it shows a fake welcome notification to the user:
Dear Customer, we’re updating your configuration and it will be ready as soon as possible.
At the same time, it hides an icon and starts background services to hide further actions from the user.
||Uploading last recorded .amr audio
||Location tracking with movement detection
||GSM tracking (CID, LAC, PSC)
||Uploading all exfiltrated data
||XMPP С&C protocol (url.plus:5223)
||Registration on C&C via HTTP (url.plus/app/pro/)
Interestingly, a self-protection feature was implemented in almost every service. Since in Android 8.0 (SDK API 26) the system is able to kill idle services, this code raises a fake update notification to prevent it:
Cybercriminals have the ability to control the implant via HTTP, XMPP, binary SMS and FirebaseCloudMessaging (or GoogleCloudMessaging in older versions) protocols. Such a diversity of protocols gives the attackers more flexible control. In the latest implant versions there are 48 different commands. You can find a full list with short descriptions in the Appendix. Here are some of the most notable:
- ‘geofence’ – this command adds a specified location to the implant’s internal database and when it matches a device’s current location the malware triggers and begins to record surrounding audio.
- ”social” – this command that starts the ‘AndroidMDMSupport’ service – this allows the files of any other installed application to be grabbed. The service name makes it clear that by applications the attackers mean MDM solutions that are business-specific tools. The operator can specify a path with the database of any targeted application and server-side PHP script name for uploading.
Several hardcoded applications targeted by the MDM-grabbing command
- ‘wifi’ – this command creates a new Wi-Fi connection with specified configurations from the command and enable Wi-Fi if it is disabled. So, when a device connects to the established network, this process will be in silent and automatic mode. This command is used to connect the victim to a Wi-Fi network controlled by the cybercriminals to perform traffic sniffing and man-in-the-middle (MitM) attacks.
addWifiConfig method code fragments
- ‘camera’ – this command records a video/capture a photo using the front-facing camera when someone next unlocks the device.
Some versions of the Skygofree feature the self-protection ability exclusively for Huawei devices. There is a ‘protected apps’ list in this brand’s smartphones, related to a battery-saving concept. Apps not selected as protected apps stop working once the screen is off and await re-activation, so the implant is able to determine that it is running on a Huawei device and add itself to this list. Due to this feature, it is clear that the developers paid special attention to the work of the implant on Huawei devices.
Also, we found a debug version of the implant (70a937b2504b3ad6c623581424c7e53d) that contains interesting constants, including the version of the spyware.
Debug BuildConfig with the version
After a deep analysis of all discovered versions of Skygofree, we made an approximate timeline of the implant’s evolution.
Mobile implant evolution timeline
However, some facts indicate that the APK samples from stage two can also be used separately as the first step of the infection. Below is a list of the payloads used by the Skygofree implant in the second and third stages.
Reverse shell payload
The reverse shell module is an external ELF file compiled by the attackers to run on Android. The choice of a particular payload is determined by the implant’s version, and it can be downloaded from the command and control (C&C) server soon after the implant starts, or after a specific command. In the most recent case, the choice of the payload zip file depends on the device process architecture. For now, we observe only one payload version for following the ARM CPUs: arm64-v8a, armeabi, armeabi-v7a.
Note that in almost all cases, this payload file, contained in zip archives, is named ‘setting’ or ‘setting.o’.
The main purpose of this module is providing reverse shell features on the device by connecting with the C&C server’s socket.
Reverse shell payload
The payload is started by the main module with a specified host and port as a parameter that is hardcoded to ‘184.108.40.206’ and ‘30010’ in some versions:
Alternatively, they could be hardcoded directly into the payload code:
We also observed variants that were equipped with similar reverse shell payloads directly in the main APK /lib/ path.
Equipped reverse shell payload with specific string
After an in-depth look, we found that some versions of the reverse shell payload code share similarities with PRISM – a stealth reverse shell backdoor that is available on Github.
Reverse shell payload from update_dev.zip
At the same time, we found an important payload binary that is trying to exploit several known vulnerabilities and escalate privileges. According to several timestamps, this payload is used by implant versions created since 2016. It can also be downloaded by a specific command. The exploit payload contains following file components:
||Sqlite3 tool ELF
||Sqlite3 database with supported devices and their constants needed for privilege escalation
‘device.db’ is a database used by the exploit. It contains two tables – ‘supported_devices’ and ‘device_address’. The first table contains 205 devices with some Linux properties; the second contains the specific memory addresses associated with them that are needed for successful exploitation. You can find a full list of targeted models in the Appendix.
Fragment of the database with targeted devices and specific memory addresses
If the infected device is not listed in this database, the exploit tries to discover these addresses programmatically.
After downloading and unpacking, the main module executes the exploit binary file. Once executed, the module attempts to get root privileges on the device by exploiting the following vulnerabilities:
CVE-2014-3153 (futex aka TowelRoot)
After an in-depth look, we found that the exploit payload code shares several similarities with the public project android-rooting-tools.
Decompiled exploit function code fragment
run_with_mmap function from the android-rooting-tools project
As can be seen from the comparison, there are similar strings and also a unique comment in Italian, so it looks like the attackers created this exploit payload based on android-rooting-tools project source code.
Busybox is public software that provides several Linux tools in a single ELF file. In earlier versions, it operated with shell commands like this:
Stealing WhatsApp encryption key with Busybox
Actually, this is not a standalone payload file – in all the observed versions its code was compiled with exploit payload in one file (‘poc_perm’, ‘arrs_put_user’, ‘arrs_put_user.o’). This is due to the fact that the implant needs to escalate privileges before performing social payload actions. This payload is also used by the earlier versions of the implant. It has similar functionality to the ‘AndroidMDMSupport’ command from the current versions – stealing data belonging to other installed applications. The payload will execute shell code to steal data from various applications. The example below steals Facebook data:
All the other hardcoded applications targeted by the payload:
||LINE: Free Calls & Messages
Upon receiving a specific command, the implant can download a special payload to grab sensitive information from external applications. The case where we observed this involved WhatsApp.
In the examined version, it was downloaded from:
The payload can be a .dex or .apk file which is a Java-compiled Android executable. After downloading, it will be loaded by the main module via DexClassLoader api:
As mentioned, we observed a payload that exclusively targets the WhatsApp messenger and it does so in an original way. The payload uses the Android Accessibility Service to get information directly from the displayed elements on the screen, so it waits for the targeted application to be launched and then parses all nodes to find text messages:
Note that the implant needs special permission to use the Accessibility Service API, but there is a command that performs a request with a phishing text displayed to the user to obtain such permission.
We have found multiple components that form an entire spyware system for the Windows platform.
||Main module, reverse shell
||Sending exfiltrated data
||Surrounding sound recording by mic
||Skype call recording to MP3
All modules, except skype_sync2.exe, are written in Python and packed to binary files via the Py2exe tool. This sort of conversion allows Python code to be run in a Windows environment without pre-installed Python binaries.
msconf.exe is the main module that provides control of the implant and reverse shell feature. It opens a socket on the victim’s machine and connects with a server-side component of the implant located at 220.127.116.11:6500. Before connecting with the socket, it creates a malware environment in ‘APPDATA/myupd’ and creates a sqlite3 database there – ‘myupd_tmp\\mng.db’:
CREATE TABLE MANAGE(ID INT PRIMARY KEY NOT NULL,Send INT NOT NULL, Keylogg INT NOT NULL,Screenshot INT NOT NULL,Audio INT NOT NULL);
INSERT INTO MANAGE (ID,Send,Keylogg,Screenshot,Audio) VALUES (1, 1, 1, 1, 0 )
Finally, the malware modifies the ‘Software\Microsoft\Windows\CurrentVersion\Run’ registry key to enable autostart of the main module.
The code contains multiple comments in Italian, here is the most noteworthy example:
“Receive commands from the remote server, here you can set the key commands to command the virus”
Here are the available commands:
||Change current directory to specified
||Close the socket
||Execute received command via Python’s subprocess.Popen() without outputs
||Upload specified file to the specified URL
||Download content from the specified URLs and save to specified file
||Dump file structure of the C: path, save it to the file in json format and zip it
|Enable/disable screenshot module. When enabled, it makes a screenshot every 25 seconds
|Enable/disable keylogging module
|Enable/disable surrounding sounds recording module
||Send components status to the C&C socket
||Execute received command via Python’s subprocess.Popen(), output result will be sent to the C&C socket.
All modules set hidden attributes to their files:
||Exfiltrated data format
||%Y%m%d-%H%M%S_filesystem.zip (file structure dump)
||%d%m%Y%H%M%S.wav (surrounding sounds)
(skype calls records)
Moreover, we found one module written in .Net – skype_sync2.exe. The main purpose of this module is to exfiltrate Skype call recordings. Just like the previous modules, it contains multiple strings in Italian.
After launch, it downloads a codec for MP3 encoding directly from the C&C server:
The skype_sync2.exe module has a compilation timestamp – Feb 06 2017 and the following PDB string:
network.exe is a module for submitting all exfiltrated data to the server. In the observed version of the implant it doesn’t have an interface to work with the skype_sync2.exe module.
network.exe submitting to the server code snippet
We found some code similarities between the implant for Windows and other public accessible projects.
It appears the developers have copied the functional part of the keylogger module from this project.
update.exe module and Keylogger by ‘El3ct71k’ code comparison
update.exe module and Xenotix Python Keylogger code comparison
‘addStartup’ method from msconf.exe module
‘addStartup’ method from Xenotix Python Keylogger