Loading page
Skip to main content

Setting up own mongodb honeypot to lure attackers

Editorial date
Editorial date
Blog Post views
...
Reading time
5 min read
Setting up own mongodb honeypot to lure attackers

TL;DR

I deployed T-Pot 24.04.1 (Hive) on AWS EC2 Instance t2.xlarge (8GB RAM, 80GB), exposed an unauthenticated MongoDB with seeded data, and left it open to the internet.

What happened:

  • Port 27017 was discovered within minutes
  • 158 attacks from 5 IPs in the first 24 hours
  • SSH brute force started almost immediately
  • One attacker detected the honeypot and wiped logs

After 3 days:

  • 1,000+ total attacks
  • SSH brute force dominated (Cowrie)
  • MongoDB scans: 52 attempts (Dionaea)
  • Elasticsearch probes: negligible

Network difference:

  • Public subnet → 1,000+ attacks
  • Private subnet → 1 hit

Same setup. Same rules. Only difference: subnet.

Elastic Stack dashboard showing seven days of honeypot events and attack volume

Elastic (ELK) dashboard after seven days on the public subnet honeypot

What is a Honeypot?

A honeypot is a system that pretends to be real and waits to get attacked.

You put it on the internet, you watch who hits it, and you learn exactly what attackers are doing right now. not what some outdated report claims.

Exposed Mongo databases get ransomed constantly. Misconfigured, no auth, public internet. this mistake is common and attackers rely on it. So I built the bait.

The Setup

Two EC2 instances with identical specs. Only difference: subnet.

This was intentional. the subnet difference ended up being the most interesting finding of the whole experiment.

Both machines:

  • t2.xlarge. 4 vCPU, 8 GB RAM
  • 80GB EBS volume
  • Ubuntu 24.04 LTS
  • T-Pot 24.04.1 (Hive edition)

Machine 1. Public subnet. Direct internet exposure. This is the honeypot. Machine 2. Private subnet. Behind NAT. No inbound internet route.

Public and private subnet honeypot architecture

Diagram of two identical EC2 instances running T-Pot, one in a public subnet and one in a private subnet

Identical T-Pot hosts: public vs private subnet

Why t2.xlarge?

I tried t2.medium first (4GB RAM). It couldn't finish starting all containers.

T-Pot runs 30+ Docker containers. Elasticsearch alone wants 2GB. Minimum you need 8GB RAM & 80 GB Storage otherwise it won’t even start properly.

Creating the User

T-Pot's installer is opinionated about one thing: don't run it as root, and don't name your user tpot (it creates that system user itself).

adduser devops
usermod -aG sudo devops
su - devops

That's it. Simple. I made the mistake of naming the user tpot first.

The Ansible playbook rejected it immediately with Reserved username 'tpot' detected.

Wasted 20 minutes. Don’t make this mistake.

Installing T-Pot

No manual Docker install. The script handles everything.

git clone https://github.com/telekom-security/tpotce
cd tpotce
./install.sh

It installs Docker, pulls all images, configures systemd, sets up iptables rules, and moves SSH to port 64295. Port 22 becomes a Cowrie honeypot.

Terminal output while T-Pot install pulls Docker images for honeypot services

T-Pot installer

When it asks which edition:

(H)ive   - Full install, all honeypots, ELK stack
(S)ensor - Distributed sensor, no UI
M(i)ni   - Lightweight, fewer honeypots
...

Pick H.

Mini doesn't include Dionaea. which is the only honeypot in T-Pot that covers MongoDB. Learned this the hard way after a full install, 20 minutes of debugging, and a reinstall.

After reboot:

  • SSH → port 64295
  • Web UI → https://your-ip:64297

Summary of T-Pot ports after install: SSH moved to 64295 and web dashboard on 64297

T-Potweb UI

What Dionaea Is (and What It Isn't)

Dionaea is T-Pot's database honeypot.

It emulates MongoDB on port 27017, alongside SMB, FTP, MySQL, MSSQL, and others. all in one container.

The catch: Dionaea emulates MongoDB wire protocol version 5. That's MongoDB ~3.6 era.

Modern clients like Compass or mongosh require at least version 8.

# This is what Compass returns when connecting to Dionaea
Server at 13.206.68.203:27017 reports maximum wire version 5,
but this version of the Node.js Driver requires at least 8

So Compass rejects it. But automated scripts?

They don't care.

Most MongoDB attack bots use older drivers or raw socket connections. They connect, try to enumerate databases, and move on. Dionaea logs all of it.

MongoDB Compass connection error: server reports maximum wire protocol version 5 while the driver requires version 8 or higher

For basic scanning detection, Dionaea works fine. But it can't serve real queryable data. An attacker who actually connects and runs db.users.find() gets nothing back. a decent attacker would spot this instantly.

That's where the next part comes in.

Adding a Real MongoDB. The Convincing Part

Here’s where it gets real.

I wanted attackers to see real data. Collections. Documents.

The kind of thing that makes someone think they've hit a misconfigured database.

Remove port 27017 from Dionaea. Then add a real MongoDB 6.0 container:

# Add to networks section
mongodb_local:

# Add to services section
mongodb_honeypot:
  container_name: mongodb_honeypot
  image: mongo:6.0
  restart: always
  networks:
    - mongodb_local
  ports:
    - "27017:27017"
  volumes:
    - /data/mongodb/data:/data/db
    - /data/mongodb/log:/var/log/mongodb
    - /home/devops/mongodb-honeypot/seed.js:/docker-entrypoint-initdb.d/seed.js:ro
  environment:
    - MONGO_INITDB_DATABASE=companydb
  command: mongod --bind_ip_all --logpath /var/log/mongodb/mongod.log --logappend

No auth. That's the bait.

The seed file runs automatically on first start.

MongoDB executes anything in /docker-entrypoint-initdb.d/ on initialization:

db = db.getSiblingDB("companydb");

db.users.insertMany([
  {
    name: "Rahul Sharma",
    email: "rahul.sharma@gmail.com",
    password: "Rahul@1234",
    phone: "+91-9876543210",
    city: "Hyderabad",
    role: "admin",
  },
  // ... more users
]);

db.finance_transactions.insertMany([
  {
    txn_id: "TXN2024031501",
    user: "rahulsharma438@gmail.com",
    amount: 125000,
    method: "NEFT",
    bank: "HDFC Bank",
    account_no: "HDFC0012345678",
    ifsc: "HDFC0001234",
  },
  // ... more transactions
]);

Indian names. Indian phone numbers. UPI IDs. IFSC codes. NEFT/RTGS transactions. Ad spend data with INR budgets. It looks like a misconfigured Indian Company database.

Start it:

Docker Desktop or CLI showing the mongodb_honeypot container running alongside T-Pot services

MongoDB honeypot container

cd ~/tpotce

# Restart Dionaea without 27017
sudo docker compose up -d --no-deps dionaea

# Start the real MongoDB
sudo docker compose up -d mongodb_honeypot

Test from your machine:

mongosh "mongodb://your-ec2-ip:27017"
use companydb or random company name
db.users.find()
# Returns all your fake users

mongosh session listing seeded fake user documents from the companydb database

What Actually Happened

First 24 Hours

158 attacks. 5 unique IPs.

Port 27017 was found by automated scanners within minutes of going live.

These bots continuously scan the internet, and exposed services like this end up indexed on platforms like Shodan and Censys within minutes.

Then this happened: One attacker connected, poked around, realized it was a honeypot. and wiped the logs before leaving.

That's not automated. That's someone who knew what they were looking at, tried to cover their tracks, and left.

The fake data was convincing enough to trigger a human response.

Honeypot attack sequence

Honeypot dashboard metrics for the first day: attack counts and sources after going live

First 24 hours on the dashboard

After 3 Days

Honeypot statistics after three days: total hits and breakdown by honeypot service

Three-day aggregate attack statistics

Hits per honeypot after three days (public subnet)

HoneypotHitsWhat it caught
Cowrie~1,000SSH/Telnet brute force
Dionaea52MongoDB protocol scans
ElasticPot3Elasticsearch exploits

Top attacker countries: Hong Kong, US, China, Australia, India.

Top ASN: UCLOUD with 956 hits. One IP. 956 attempts.

Map and charts of attack origins by country and autonomous system number

Geography and ASN of incoming attacks

Password tagcloud from Cowrie SSH brute force:

Word cloud of passwords attempted against the Cowrie SSH honeypot, including common defaults

Cowrie: most-tried SSH passwords

debian. ubuntu. linux. 123456. raspberry.

If any of systems use these passwords they already have a problem.

These aren't targeted guesses. They're credential lists from leaked databases, running automatically against every IP on the internet.

Honeypot UI summarizing attack volume across Cowrie, Dionaea, and related services

What actually matters

Public subnet: 1,000+ attacks in 3 days

Private subnet: 1 attack

Only difference: network placement.

Side-by-side comparison of attack counts: thousands of hits on the public subnet honeypot versus almost none on the private subnet

Public subnet vs private subnet

This isn't about firewall rules. Security groups are table stakes. The actual protection comes from having no inbound internet route at all. Private subnet databases literally cannot be reached from the internet.

The public subnet database gets found in minutes and becomes searchable through platforms like Shodan. It doesn't matter how good your firewall rules are if you misconfigure one rule. and people do misconfigure rules.

If RDS, MongoDB, Redis, or any database is in a public subnet, it's one security group mistake away from being the next ransom victim.

The Bad Practice

People put databases in public subnets and rely entirely on security groups to block access.

Then:

  • Open 0.0.0.0/0 on port 27017 to "test something quickly"
  • Forget to revert it
  • Come back to a ransom note replacing their data

The attackers aren't even human. it's fully automated scripts.

The fake data we seeded was designed to look like exactly the kind of thing that would get ransomed: user PII, financial transaction data, credentials.

The attacker who identified it as a honeypot and wiped logs was doing exactly what they'd do on a real target. and they would have found 980,000 INR transactions, HDFC account numbers, and admin credentials if it were real.

Takeaway

  • Put databases in private subnets. Always.
  • No public IP. No inbound route. No exceptions.
  • Use bastion hosts / SSM, not direct access
  • Treat security groups as secondary defense, not primary

If your database is in a public subnet, it’s not “secured”.
It’s just waiting to be discovered.

This isn’t targeted.

Everything is scanned continuously.
Expose MongoDB on 27017 and it gets found, indexed, and probed within minutes.

The internet is scanning every port, all the time. Yours included. 🍯

Putting a database in a public subnet and relying on firewall rules is like:

locking your front door and removing the walls

It works.
Right until it doesn’t.

And when it doesn’t, it’s not subtle.