Skip to navigation

Threat reportMalwareTL-2026-2857

Cling IoT botnet masquerades as Google STUN traffic for C2, exploiting Realtek Jungle SDK CVE-2021-35394

criticalACTIVE

Nozomi Networks Labs describes Cling, a MIPS embedded-Linux botnet that spreads through CVE-2021-35394 (Realtek Jungle SDK UDPServer/MP Daemon command injection) and hides its command channel in STUN Binding traffic, carrying operator commands in the 12-byte transaction ID and spoofing the source address of stun.l.google.com.

Key facts

IdentifiersTL-2026-2857, CVE-2021-35394, CVE-2014-8361, CVE-2023-26801, CVE-2024-3721, CVE-2025-34037, CVE-2016-10372, CVE-2023-41011, CVE-2016-20016, CVE-2023-1389, CVE-2024-7029, CVE-2025-34035, CVE-2024-23625, CVE-2024-23624, CVE-2019-17621, CVE-2019-7256, CVE-2021-36380, CVE-2024-21887, CVE-2023-46805, CVE-2024-46048, CVE-2024-35340, CVE-2024-32314, CVE-2024-32292, CVE-2024-32281, CVE-2022-35555, CVE-2022-26289, CVE-2022-36553, CVE-2022-37055, CVE-2024-10915, CVE-2024-10914, CVE-2026-36356, CVE-2025-67038, CVE-2026-87827
Also known asCling
SeverityCRITICAL
CVSS9.8 (CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H)
Exploitation statusLinked CVE exploited in the wild; Listed in CISA KEV: CVE-2021-35394 (added ), CVE-2014-8361 (added ), CVE-2023-1389 (added ), CVE-2019-17621 (added ), CVE-2019-7256 (added ), CVE-2021-36380 (added ) and 4 more
EPSSCVE-2023-1389: 99.9% (99.9th percentile); CVE-2024-21887: 99.9% (99.9th percentile); CVE-2023-46805: 99.9% (99.9th percentile)
Affected productsRealtek Jungle SDK (MP Daemon / UDPServer); LB-LINK Routers (CVE-2023-26801); TBK DVR (CVE-2024-3721); Linksys Routers (CVE-2025-34037); Eir D1000 router (CVE-2016-10372); FiberHome SR1041F (CVE-2023-41011); MVPower CCTV DVR (CVE-2016-20016)
Weaknesses (CWE)CWE-78, CWE-77, CWE-94, CWE-306
MalwareCling
MITRE ATT&CK techniquesT1001 Data Obfuscation, T1001.003 Protocol or Service Impersonation, T1036.003 Rename Legitimate Utilities, T1037.004 RC Scripts, T1046 Network Service Discovery, T1057 Process Discovery, T1059.004 Unix Shell, T1090 Proxy, T1095 Non-Application Layer Protocol, T1105 Ingress Tool Transfer, T1190 Exploit Public-Facing Application, T1210 Exploitation of Remote Services and 6 more in the attack flow
MotivationUNKNOWN
Target sectorstelecoms, consumer-iot, surveillance, critical-infrastructure
Target regionsGlobal
Detection rules9 (1 shown free)
Sources9 cited sources

Detection coverage

As of 2026-10-08, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-2857 across Splunk SPL, Microsoft KQL and Sigma, covering 57 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.

Free detection: Cling — .cling Persistence Copy and Init Script Modification

Detects the Cling persistence mechanism: copies of the bot at /root/.cling and /usr/local/bin/.cling, and modification of /etc/inittab, /etc/init.d/rcS or /etc/rc.d/rc.boot with references to .cling so the bot restarts at boot on SysV and BusyBox init layouts.

  • Languages: Splunk SPL, Microsoft KQL, Sigma
  • ATT&CK: T1037.004
  • Data sources: auditd, osquery, file_integrity_monitoring, syslog, DeviceFileEvents, DeviceProcessEvents
  • Severity: high
  • Confidence: high

Splunk SPL

```Cling persistence: .cling copies plus init script edits referencing .cling```
index=linux (sourcetype=linux:audit OR sourcetype=auditd OR sourcetype=osquery:results OR sourcetype=fim OR sourcetype=syslog) (".cling" OR "/etc/inittab" OR "/etc/init.d/rcS" OR "/etc/rc.d/rc.boot") | eval path_hit=coalesce(file_path, path, name, object_path, _raw) | eval cling_artifact=if(match(path_hit, "(?i)(^|/)[.]cling($|[^a-z0-9_])"), 1, 0) | eval init_script=if(match(path_hit, "(?i)/etc/(inittab|init[.]d/rcS|rc[.]d/rc[.]boot)"), 1, 0) | where cling_artifact=1 OR (init_script=1 AND match(_raw, "(?i)[.]cling")) | stats count min(_time) as first_seen max(_time) as last_seen values(path_hit) as paths values(user) as users by host | convert ctime(first_seen) ctime(last_seen)

Microsoft KQL

let ClingFiles = DeviceFileEvents
| where Timestamp > ago(24h)
| where FileName =~ '.cling'
| project Timestamp, DeviceName, Signal = 'cling-file', Detail = strcat(FolderPath, '/', FileName), Process = InitiatingProcessCommandLine;
let InitAppend = DeviceProcessEvents
| where Timestamp > ago(24h)
| where ProcessCommandLine contains '.cling'
| where ProcessCommandLine has_any ('/etc/inittab', '/etc/init.d/rcS', '/etc/rc.d/rc.boot')
| project Timestamp, DeviceName, Signal = 'init-script-append', Detail = ProcessCommandLine, Process = InitiatingProcessCommandLine;
union ClingFiles, InitAppend
| order by Timestamp desc

Sigma

title: Cling IoT Botnet Persistence Copy Named .cling
id: 7a2d3b5f-9c4e-4f68-8b21-3d8e6f9a2b02
status: experimental
description: Detects creation of a hidden file named .cling, the persistence copy Cling writes to /root/.cling and /usr/local/bin/.cling before wiring it into init scripts.
references:
  - https://www.nozominetworks.com/blog/a-stunning-disguise-cling-malware-masquerades-as-google-
author: Threadlinqs Team
date: 2026-10-03
tags:
  - attack.persistence
  - attack.t1037.004
  - cve.2021-35394
logsource:
  category: file_event
  product: linux
detection:
  selection:
    TargetFilename|endswith: '/.cling'
  condition: selection
falsepositives:
  - Unrelated software that happens to use a hidden file named .cling
level: high

Known false positives:

  • Unrelated software using a hidden file literally named .cling
  • Legitimate administrative edits to init scripts (only alert when .cling is referenced)

Tuning notes: Most embedded devices cannot run an agent; forward auditd or FIM events from Linux gateways, or use firmware integrity checks on exported device filesystems. The init-script branch needs command-line or content visibility; without it, rely on the .cling filename match.

8 more detection rules (Splunk SPL, Microsoft KQL, Sigma) · unlock with TL Creds. Sign up free to unlock the detection rules or query them with the Threadlinqs MCP server (Purple)
  • Cling — BusyBox wget Fetch of mipsel Loader / [indicator] Tag · unlock with TL Creds
  • Cling — wget Binary Hijack ([indicator] / [indicator] Artifacts) · unlock with TL Creds
  • Cling — Internal Host Fan-Out to Many Public STUN Servers · unlock with TL Creds
  • Cling — IoT Scanner Fan-Out to Many Public IPv4 Hosts · unlock with TL Creds
  • Cling — Outbound DDoS Flood from an Internal IoT Device · unlock with TL Creds
  • Cling — Traffic to Operator STUN Server and Known Loader Hosts · unlock with TL Creds
  • Cling / CVE-2021-35394 — Inbound UDP 9034 Realtek UDPServer 'orf;' Command Injection · unlock with TL Creds
  • Cling — STUN Binding Request with All-Zero Transaction ID (Covert C2) · unlock with TL Creds

Indicators of compromise

As of 2026-10-08, Threadlinqs Intelligence tracks 57 indicators of compromise for TL-2026-2857.

Sample indicators (3 of 57)

  • IP address: 118.45.196[.]225
  • File hash: 3b0ac6aaabb3bf8058ca14f9c8ccc613cfa3ea71
  • IP address: 120.193.219[.]210

Defanged, TLP:CLEAR. Full set of 57 indicators: unlock with TL Creds.

57 indicators of compromise · unlock with TL Creds. Compare plans

Technical analysis

It persists through init scripts and by replacing wget, and supports scanning, proxying, TCP tunneling and DDoS.

Cling was found by Nozomi Networks Labs while investigating exploitation of CVE-2021-35394, a Realtek Jungle SDK flaw (versions 2.0 through 3.4.14B) in the diagnostic 'MP Daemon', usually compiled as the 'UDPServer' binary listening on UDP 9034. Attackers send UDP datagrams beginning with 'orf;' followed by shell commands. The commands use BusyBox wget to fetch a MIPS (mipsel) binary from a loader host and execute it with an infection-method tag (for example 'realtek.selfrep') as the first argument.

On the device, the malware binds a socket with SO_REUSEADDR to TCP/UDP port 33957 as a single-instance check and exits if the bind fails. It copies itself to /root/.cling and /usr/local/bin/.cling and appends references to /etc/inittab, /etc/init.d/rcS and /etc/rc.d/rc.boot, which covers SysV and BusyBox init layouts. It also hijacks wget: the legitimate binary is moved to wget.r, its original path is stored in wget.p, and the malware takes its place, so later wget invocations re-run or reinstall the bot.

The C2 channel abuses STUN (RFC 8489). Every ~5 seconds the bot sends STUN Binding Requests to 13 hardcoded public STUN servers using an all-zero transaction ID instead of a random one. It then sends a non-standard registration datagram containing the infection tag and its mapped external ports to each server; compliant servers ignore it. The bot then listens on the mapped ports for UDP packets whose 12-byte STUN transaction ID field encodes operator commands. Nozomi identified 145.249.115.184:3478 as an operator-controlled STUN server because it answered with all-zero transaction IDs instead of echoing requests. Command packets were seen arriving from 74.125.250.129 (stun.l.google.com); Nozomi assesses this as UDP source-address spoofing, based on TTL differences.

Eight commands are encoded in the transaction ID: execute (download a payload from IP:port and pass it to system(3)), scan-and-exploit across IPv4, stop scanner, start TCP tunnel, stop TCP tunnel, proxy relay (bidirectional forwarding to a relay server), stop proxy, and flood (DDoS for a set duration). Observed DDoS targets included a South Korean ISP host (112.151.157.222:8080), a University of Chicago host (192.170.240.137:53) and two Minecraft servers (23.81.40.193:25565, 147.185.221.129:25565). Besides CVE-2021-35394, the bot embeds exploits for CVE-2014-8361 (Realtek SDK), CVE-2023-26801 (LB-LINK routers), CVE-2024-3721 (TBK DVR), CVE-2025-34037 (Linksys), CVE-2016-10372 (Eir D1000), CVE-2023-41011 (FiberHome SR1041F) and CVE-2016-20016 (MVPower CCTV DVR). No threat actor attribution is stated in the sources.

CVE-2021-35394 (CVSS 3.1 base 9.8, CWE-78) was published 2021-08-16 and is in the CISA KEV catalog. Unit 42 reported 134 million exploit attempts between August and December 2022 and counted about 190 affected device models from 66 vendors, with Mirai, Gafgyt, Mozi and RedGoBot using it. Defenders should hunt for .cling files, wget.r/wget.p artifacts, modified init scripts, repeated STUN Binding Requests with all-zero transaction IDs, and non-STUN UDP payloads sent to STUN endpoints. Source reputation alone is insufficient because the C2 spoofs a Google address.

MITRE ATT&CK techniques by tactic:

Command and Control

T1001 Data Obfuscation; T1001.003 Protocol or Service Impersonation; T1090 Proxy; T1095 Non-Application Layer Protocol; T1105 Ingress Tool Transfer; T1572 Protocol Tunneling

Defense Evasion

T1036.003 Rename Legitimate Utilities; T1562.001 Disable or Modify Tools; T1564.001 Hidden Files and Directories

Persistence

T1037.004 RC Scripts

Discovery

T1046 Network Service Discovery; T1057 Process Discovery

Execution

T1059.004 Unix Shell

Initial Access

T1190 Exploit Public-Facing Application

Lateral Movement

T1210 Exploitation of Remote Services

Impact

T1498.001 Direct Network Flood

stealth

T1574 Hijack Execution Flow

Resource Development

T1584.008 Compromise Infrastructure: Network Devices

Remediation

Patches

  • Apply vendor updates for Realtek Jungle SDK CVE-2021-35394 (fixed after v3.4.14B)
  • Patch the embedded-exploit targets: CVE-2014-8361, CVE-2023-26801, CVE-2024-3721, CVE-2025-34037, CVE-2016-10372, CVE-2023-41011, CVE-2016-20016

Immediate actions

  • Hunt embedded devices for /root/.cling, /usr/local/bin/.cling, and wget.r / wget.p files beside wget in /bin, /sbin, /usr/bin, /usr/sbin and /usr/local/bin
  • Alert on STUN Binding Requests with an all-zero transaction ID and on non-STUN UDP datagrams sent to STUN endpoints
  • Block egress to 145.249.115.184 and the loader hosts 118.45.196.225, 120.193.219.210, 58.211.144.243 and 121.32.243.81
  • Block inbound UDP 9034 from the internet to Realtek-based devices

Workarounds

  • Replace or retire end-of-life devices that cannot be patched
  • Audit /etc/inittab, /etc/init.d/rcS and /etc/rc.d/rc.boot for unexpected entries and reflash compromised devices from known-good firmware

Longer-term hardening

  • Remove unnecessary internet-facing access to routers, DVRs, access points and embedded appliances
  • Segment vulnerable IoT and edge devices into isolated zones and treat them as network entry points
  • Baseline asset connectivity and alert on deviations; inspect protocol behavior instead of relying on destination reputation alone

Affected and fixed versions

  • Realtek — Jungle SDK (MP Daemon / UDPServer)
    Vulnerable versions: 2.0 through 3.4.14B
    Fixed in: Apply vendor patch per Realtek advisory
  • LB-LINK — Routers (CVE-2023-26801)
  • TBK — DVR (CVE-2024-3721)
  • Linksys — Routers (CVE-2025-34037)
  • Eir — D1000 router (CVE-2016-10372)
  • FiberHome — SR1041F (CVE-2023-41011)
  • MVPower — CCTV DVR (CVE-2016-20016)

Timeline

Last updated · 3 updates · revalidated 3× · latest source

  • CVE-2021-35394 (Realtek Jungle SDK MP Daemon/UDPServer command injection and memory corruption, CVSS 9.8) is published in NVD.
  • CVE-2021-35394 is listed in CISA's Known Exploited Vulnerabilities catalog (date per NVD KEV data).
  • Start of the August-December 2022 window in which Unit 42 observed 134 million exploit attempts against CVE-2021-35394, across roughly 190 device models from 66 vendors.
  • Mass exploitation of CVE-2021-35394 begins; RedGoBot first observed. Unit 42 later counts 134 million attempts from August to December 2022, with Mirai, Gafgyt and Mozi also using the flaw.
  • End of the Unit 42 observation window for mass CVE-2021-35394 exploitation (134 million attempts August-December 2022), establishing the vulnerable Realtek population that Cling later targets.
  • Exploitation of CVE-2021-35394 delivering the Cling botnet observed starting around this date.
  • Date recorded in the YARA rule Nozomi published for Cling samples, indicating the samples were being tracked before the 2026-10-01 disclosure.
  • Nozomi Networks Labs publishes its analysis of Cling: STUN transaction-ID C2, spoofed stun.l.google.com source, wget hijack and eight embedded exploits.
  • CyberPress also covers Cling, confirming 13 hardcoded STUN servers queried about every 5 seconds, all-zero transaction IDs and the suspected colluding server locked-ioc-23:3478.
  • GBHackers reports on Cling, including the IOC set and MITRE ATT&CK mapping from the Nozomi research.
  • FortiGuard documents three ClingSTUN evolutions by download host: locked-ioc-25 (about two days, single exploit CVE-2022-36553), locked-ioc-30 (multi-exploit, 24 STUN servers, 50% STUN-success threshold) and locked-ioc-26 (current; 13 STUN endpoints, 100% threshold, adds mount/process enumeration). Observation dates are not stated in the report.
  • The Hacker News reports Realtek Jungle SDK exploit attempts delivering Cling with STUN-based C2.
  • Fortinet FortiGuard Labs publishes 'ClingSTUN Linux Backdoor Abuses Public STUN Infrastructure', documenting 20+ entry-point CVEs, 24 STUN endpoints, new samples and additional ARM/x86/PowerPC builds.

Update history

Sources cited

Other reports on CVE-2021-35394

Other reports on CVE-2014-8361

Previous and next in malware

Drafted by the Threadlinqs research pipeline from 9 sources; published ; last updated .

Threadlinqs Intelligence — Real-Time Threat Detection Platform

// threat_feed
$ sort --newest
Showing all threats

Live intelligence console

Threat level

Latest Threats