container-escape-techniques

An offensive-security playbook for breaking out of Docker containers, LXC, and Kubernetes pods to the underlying host, intended for authorized penetration testing. It applies once an operator is inside a container and needs to reach the host through misconfiguration or capability abuse, and it routes to related privilege-escalation, Kubernetes-pentesting, and security-bypass material as needed. It opens with container detection using /proc/1/cgroup, /.dockerenv, mountinfo, and tools such as amicontained, deepce, and CDK.

Escape paths are grouped by root cause. Privileged containers allow mounting the host disk (for example /dev/sda1) and using nsenter --target 1 to enter host namespaces for an effective host shell, and with hostPID the host filesystem is reachable through /proc/1/root. Capability abuse covers CAP_SYS_ADMIN for mount-based escapes, CAP_SYS_PTRACE for process injection, CAP_NET_ADMIN for host network manipulation, and CAP_DAC_READ_SEARCH via the shocker open_by_handle_at technique. A mounted Docker socket enables creating a privileged container through the CLI or the raw HTTP API with curl.

Further sections detail the cgroup v1 release_agent escape for CAP_SYS_ADMIN with cgroup v1, considerations for cgroup v2 and eBPF, and namespace escapes via unshare and shared PID namespaces. Runtime vulnerabilities include runc CVE-2019-5736, containerd CVE-2020-15257, and cgroups CVE-2022-0492. A Kubernetes pod-escape matrix maps dangerous pod specs (hostPID, hostNetwork, hostPath mounting root, privileged, and over-privileged service-account tokens) to their escapes, and a tool table lists deepce, CDK, amicontained, PEIRATES, and BOtB. Presented factually, the same knowledge informs defenders hardening container platforms.