WTF my container just spawned a shell!

Jorge Salamero Sanz@bencerillo

WTF my container just spawned a shell!

Sysdig

Open Source system troubleshooting

with native container support(htop, vmstat, netstat, lsof, tcpdump…)

Monitoring, alerting,

troubleshooting tool for

Docker, Kubernetes, Mesos,

RancherOS, GCE, ECS

Securing Microservices

Defense in depth Microservices

Single responsibility principle:

• Principle of least privileges

• Principle of least surprise

• Principle of least access

Container Security Techniques

• New workflows (docker pull vs previous apt update/yum update)

• CI/CD Pipeline (Jenkins, Openshift, etc: rebuilds, rolling updates)

• Access management, logging & auditing (Kubernetes, Openshift:

user roles, namespaces, centralized logging, authn & authz, etc)

• Trust boundaries (do you trust entire Docker Hub? Image signing, Docker Notary)

• Simplify container OS, tons of useless stuff (RancherOS, Atomic,

CoreOS, Alpine Linux, etc)

• Network security (Weave, etc)

• Secret management (Vaults instead of environment vars, etc)

Docker Security Techniques

• Drop privileges

• Limit capabilities

• Do not run as root! containers are not VMs!

• cgroups

• Limit resources share, not access restriction

• namespaces

• Useful for not complete security model

• User namespaces

• Kubernetes/Openshift security contexts

• Host security and access (Docker bench, RO access, etc)

Scanning

What are my containers doing?

• Static scanning

• Dynamic scanning

Static Scanning

Yay, this was soo easy to deploy! I Docker Hub!(your developers too, actually they were already using it :P)

uhm… wait, is someone maintaining this image?

RUN apt-get install -y wget build-essential python python-dev python-pip python-virtualenvRUN wget http://nodejs.org/dist/node-latest.tar.gzRUN tar xvzf node-latest.tar.gzRUN cd node-v* && ./configure && CXX="g++ -Wno-unused-local-typedefs" make && CXX="g++ -Wno-unused-local-typedefs" make install

• Static Scanning:

• CoreOS Clair, Docker Nautilus, Red Hat CloudForms, etc

Image Vulnerability Scanning

• Scan contents of images looking for software versions with known defects

• Container image layering can make this efficient (exploits immutable nature of images)

Ubuntu: 14.04

Apache: 2.2

Wordpress: 4.6 PHP: 7.0

Image Vulnerability Scanning

• Scan contents of images looking for software versions with known defects

• Container image layering can make this efficient (exploits immutable nature of images)

•

Ubuntu: 14.04

Apache: 2.2

Wordpress: 4.6 PHP: 7.0

Image Vulnerability Scanning

• Scan contents of images looking for software versions with known defects

• Container image layering can make this efficient (exploits immutable nature of images)

•

Ubuntu: 14.04

Apache: 2.2

Wordpress: 4.6 PHP: 7.0

Image Vulnerability Scanning

• Scan contents of images looking for software versions with known defects

• Container image layering can make this efficient (exploits immutable nature of images)

•

Ubuntu: 14.04

Apache: 2.2

Wordpress: 4.6 PHP: 7.0

Container Security Techniques

OK, no known vulnerabilities, still secure?

Containers are black boxes exposing a behaviour, is something misbehaving?

• Dynamic Scanning:

• Enforcement

• Auditing

Tools

• Basic sandboxing: seccomp

• Sandboxing with policies: seccomp-bpf

• Mandatory access control systems: SELinux, AppArmor

• System auditing: Auditd

• Behavioral monitoring: Falco

Seccomp

• Seccomp: application system call “sandboxing”

• One-way transition to a restricted state

• Limited set of system calls

• SECCOMP_MODE_STRICT: read(), write(), _exit(), and sigreturn()

• Process SIGKILLed when doing anything else

• No wide real use

Seccomp-bpf

• Security profiles per application/container:

• Create filter (BPF program ) with lists of allowed syscalls

• Each syscall matched against filter

• Multiple actions: SIGKILL, SIGSYS, failing the syscall, notify an attached tracer process to skip/change it, allow it

• Docker runs containerized process under a seccomp profile, multiple available

• Notable disallowed syscalls:

• clone (creating new namespaces)

• reboot (reboot the host)

• setns (change namespaces)

Mandatory Access Control

• SELinux or AppArmor

• Same mechanisms: kernel-level interception/filtering

• features++ && complexity++

• Above syscalls:

• Actors (process)

• Actions (read/write on files/sockets)

• Targets (files, IPs, ports)

• Mandatory while seccomp is voluntary

• But what if I only want to put some surveillance in?

Auditd

• Access monitoring system• Logger for SELinux• Rules based on files and syscalls:

# Alert whenever anyone performs an unlink() for a file below /usr/bin-a always,exit -S unlink -S unlinkat -F dir=/usr/bin -F success=1

# Watch any invocation of /usr/bin/passwd-w /usr/bin/passwd -p x -k passwd_mgmt

Sysdig Falco

An anomaly detection system built on top of the sysdig engine

Sysdig

Kernel

Docker

Container1

Container2

Container3

App Apprkt LXC

Kernel moduleInstrumentation

Event stream

OpenReadClose

ConnectReadWriteReadReadWriteClose

Dump to Disk

Filter

Analyze

Events context

• Syscalls are turned into events with context:

• the process name performing the system call• the process’s parents, grandparents, etc• the remote IP address to which the process is

communicating• the directory of the file being read/written• the current memory usage of the process

What is Sysdig Falco?

• Detects suspicious activity defined by a set of easy rules

• Uses sysdig’s flexible and powerful filtering expressions (in userspace -single point of failure-)

• Container support (Docker, Kubernetes, Mesos, etc)• Flexible notification methods• Open Source

Quick Examples

A shell is run in a container container.id != host and proc.name = bash

Overwrite system binaries fd.directory in (/bin, /sbin, /usr/bin, /usr/sbin) and write

Container namespace change evt.type = setns and not proc.name in (docker, sysdig)

Non-device files written in /dev(evt.type = creat or evt.arg.flags contains O_CREAT) and proc.name != blkid and fd.directory = /dev and fd.name != /dev/null

Process tries to access cameraevt.type = open and fd.name = /dev/video0 and not proc.name in (skype, webex)

Falco Rules

• .yaml file containing Macros, Lists, and Rules

- macro: bin_dir condition: fd.directory in (/bin, /sbin, /usr/bin, /usr/sbin)

- list: package_mgmt_binaries items: [dpkg, dpkg-preconfigu, rpm, rpmkey, yum, frontend]

- rule: write_binary_dir desc: an attempt to write to any file below a set of binary directories condition: bin_dir and evt.dir = < and open_write and not package_mgmt_procs output: "File below a known binary directory opened for writing (user=%user.name command=%proc.cmdline file=%fd.name)" priority: WARNING

Alerts and Outputs

• Events that match filter expression (rule) result in alerts• output field used to format event into alert message• Sending to:

• syslog• file• stdout• shell (e.g. mail -s "Falco Notification" alerts@example.com)

• Sysdig Cloud

Falco Demo

Further reading

● Documentation:○ https://sysdig.com/blog/selinux-seccomp-falco-technical-dis

cussion/

○ https://sysdig.com/blog/friends-dont-let-friends-curl-bash/

● http://www.sysdig.org/falco/

● https://github.com/draios/falco

Thank You!