
Cling botnet disguises C2 inside STUN protocol traffic and spoofs Google's STUN service to hide commands, exploiting CVE-2021-35394 and six other IoT flaws to spread and launch DDoS attacks. Key findings: - Cling exploits CVE-2021-35394, an RCE in the Realtek Jungle SDK diagnostic component (compiled as UDPServer) found in routers, access points, repeaters, and embedded appliances that rarely receive firmware updates. The malware also carries exploit code for six additional CVEs: 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/China Mobile), and CVE-2016-20016 (MVPower CCTV DVR), giving it a wide propagation surface across commodity networking and surveillance hardware. - The C2 design encodes operator commands inside STUN transaction IDs, sending traffic that resembles routine NAT-traversal exchanges from tools like Microsoft Teams, Zoom, and WebRTC browsers. Infected devices contact a hardcoded list of 13 STUN servers; one of them, confirmed operator-controlled at 145.249.115[.]184, receives bot registrations containing infection metadata and reachability info. - Commands appear to originate from an IP belonging to Google's stun.l[.]google[.]com service. Nozomi's analysis points to source-address spoofing through a network provider that does not validate source addresses. TTL differences between genuine STUN replies and command-bearing packets are one of the few network-level tells that something is wrong. - Observed commands included internet-wide scanning for additional vulnerable targets, TCP tunneling, proxying, and UDP/TCP flood attacks. Confirmed DDoS targets included a South Korean 🇰🇷 ISP, a University of Chicago cluster, and two Minecraft servers. - Host artifacts are concrete and huntable: malware copies named .cling, persistence entries written to init scripts, and a replaced wget binary with companion files named wget.r and wget.p. Because most compromised devices offer no EDR telemetry, these filesystem indicators and network-layer patterns are often the only evidence available. Detection priority: at the network layer, hunt for STUN Binding Requests sent at short, regular intervals with transaction IDs set to all zeros, and for non-STUN UDP datagrams directed at known STUN endpoints. At the host layer, scan internet-facing embedded devices for files named .cling, modified init scripts, and replaced wget binaries. Reputation alone cannot be trusted here as packets sourced from high-reputation infrastructure are exactly what the operator is manufacturing. The full IOC list and ATT&CK mapping are in the Nozomi Networks Labs report. #DFIR_Radar



