Docker Security workshop slides

Docker Security Workshop David Lawrence Riyaz Faizullabhoy Viktor Stanchev

Features See what features are available in Docker What do they do? How do you use them?

Understanding Look at some tools See underlying implementation details Learn best practices

Docker is additive to the security of your application ...

… even if you don’t use any of the techniques we cover

Docker aims to be Secure by Default https://docs.docker.com/engine/security/non-events/ CVE-2013-1956, 1957, 1958, 1959, 1979, CVE-2014-4014, 5206, 5207, 7970, 7975, CVE-2015-2925, 8543, CVE- 2016-3134, 3135, CVE-2014-0181, CVE-2015-3339, CVE-2014-4699, CVE-2014-9529, CVE-2015-3214, 4036, CVE-2016-0728, CVE-2016-2383

How do we think about containers?

How Docker Actually Works systemd docker Host OS alpine ubuntunginx sh nginx bash vimvi nginx

Namespaces: “what containers can see” ls -la /proc/<pid>/ns/

What’s namespaced? Cgroup CLONE_NEWCGROUP Cgroup root directory IPC CLONE_NEWIPC System V IPC, POSIX message queues Network CLONE_NEWNET Network devices, stacks, ports, etc. Mount CLONE_NEWNS Mount points PID CLONE_NEWPID Process IDs User CLONE_NEWUSER User and group IDs UTS CLONE_NEWUTS Hostname and NIS domain name

Demo using namespaces directly Create a shell process with pid and fs namespaces. $ sudo unshare -fp $ sudo unshare -fp --mount-proc

cgroups: “what containers can use” Aka “Control Groups” - limit container resources! CPU Memory PIDs

CPU 1 CPU 2 CPU 3 CPU 4 cgroups: “what containers can use”

CPU 1 CPU 2 CPU 3 CPU 4 CPU 1 CPU 1 CPU 2 cgroups: “what containers can use”

:(){ :|: & };: cgroups: “what containers can use”

git clone https://github.com/riyazdf/dockercon-workshop.git - cgroups directory NOTE: Ubuntu 15.10 does not support PID limits, but 16.04 does if you have it So DO NOT run the fork bomb unless you have another machine. Set up your AWS instance - check your email! chmod 400 <PATH_TO_FILE>/<name>.pem ssh -i <PATH_TO_FILE>/<name>.pem ubuntu@<Public DNS> Example: ssh -i riyaz.pem ubuntu@ec2-54-149...compute.amazonaws.com Hands-On Exercise

Securing Client-Engine Communications “My first 5 minutes…”

Docker Client Server Architecture docker client docker engine unix socket tcp socket

Exposing your engine to the internet Edit config at /lib/systemd/system/docker.service - ExecStart=/usr/bin/docker daemon -H fd:// + ExecStart=/usr/bin/docker daemon -H fd:// -H tcp://0.0.0.0:2376 Restart Docker $ sudo systemctl daemon-reload $ sudo systemctl restart docker

One Way TLS - Same way we trust websites: - Server cert and key on engine - CA cert on client - client authenticates Docker engine Docker client

Creating a CA # use a strong passphrase! $ openssl genrsa -aes256 -out ca-key.pem 4096 $ openssl req -new -x509 -days 365 -key ca-key.pem -sha256 -out ca.pem

Creating the daemon cert and key $ openssl genrsa -out server-key.pem 4096 $ openssl req -subj "/CN=$HOSTNAME" -sha256 -new -key server-key.pem -out server.csr $ echo subjectAltName = IP:10.10.10.20,IP:127.0.0.1 > extfile.cnf $ openssl x509 -req -days 365 -sha256 -in server.csr -CA ca.pem -CAkey ca-key.pem -CAcreateserial -out server-cert.pem -extfile extfile.cnf $ sudo cp ca.pem /root/.docker/ca.pem

Starting the daemon with the cert and key $> tree /etc/docker ├── key.json ├── server-cert.pem ├── server-key.pem $> /usr/bin/docker daemon -H tcp://0.0.0.0:2376 -H unix:///var/run/docker.sock --storage-driver aufs --tlsverify --tlscert /etc/docker/server-cert.pem --tlskey /etc/docker/server-key.pem

Trusting the daemon’s cert on the client $> tree ~/.docker ├── config.json ├── ca.pem OR $> export DOCKER_CERT_PATH=~/my_cert_directory $> tree ~/my_cert_directory ├── ca.pem

Secure by default: docker-machine docker-machine does this automatically to set up TLS for you by default!

Best practice: Mutual TLS Client also presents certificate Sends after verifying server cert Mutual authentication Client CA on daemon (engine) Docker client

Creating client cert and key $ openssl genrsa -out key.pem 4096 $ echo extendedKeyUsage = clientAuth > extfile.cnf $ openssl x509 -req -days 365 -sha256 -in client.csr -CA ca.pem -CAkey ca-key.pem -CAcreateserial -out cert.pem -extfile extfile.cnf

Trusting the client cert on the daemon $> tree /etc/docker ├── key.json ├── server-cert.pem ├── server-key.pem ├── ca.pem $> /usr/bin/docker daemon -H tcp://0.0.0.0:2376 -H unix:///var/run/docker.sock --storage-driver aufs --tlsverify --tlscert /etc/docker/server-cert.pem --tlskey /etc/docker/server-key.pem --tlscacert /etc/docker/ca.pem

Using the client certs on the client $> tree ~/.docker ├── config.json ├── ca.pem ├── cert.pem ├── key.pem OR $> export DOCKER_CERT_PATH=~/my_cert_directory $> tree ~/my_cert_directory ├── ca.pem ├── cert.pem ├── key.pem

Securing Engine-Registry Communications

What’s in an image: The Layered Filesystem

What is a layered filesystem? Combine multiple directories to look like a single filesystem Tombstoning/whiteout files to delete files from lower layers

Supported Implementations Aufs Btrfs OverlayFS Devicemapper ...

Best practice: “minimal” base images alpine ~ 2 MB from hub (1 layer!) musl libc and busybox ubuntu ~ 50 MB from hub

Best practice: verify content RUN apt-key adv --keyserver hkp://keyserver.ubuntu.com:80 --recv-keys BBEBDCB318AD50EC6865090613B00F1FD2C19886 && echo deb http://repository.spotify.com stable non-free | sudo tee /etc/apt/sources.list.d/spotify.list

Best practice: read only containers $ docker run it --rm --read-only alpine sh Mounts the container’s FS as read-only

Best practice: read-only Volumes -v /data:/data:ro

Common mistake: mount host location as writable $ docker run it --rm -v /:/host alpine sh

Best practice: minimal, read-only mounts $ docker run it --rm -v /subdir/we/need:/dir:ro alpine sh

Isolate services Control which services can talk to which other services ● Easier to audit

Links (legacy) Allow 2 specific containers to talk to each other. ● Brittle: does not survive container restarts docker run -d --name db mysql:latest docker run -d --link db wordpress

Network Namespace docker network create my_app docker run -it --rm --net=my_app alpine sh Links are dynamic, can be created to not yet created containers.

Best practice: Use Multiple Networks nginx my_app Internal service RethinkDB Proxy RethinkDB RethinkDB RethinkDB load_balancer_net app_net db_net

Common Mistake: --net=host Container can see ALL network traffic, including traffic on docker virtual networks

Common Mistake: ports exposed on host Unnecessary Creates conflicts

Best practice: Mutual TLS Implementation detail: use mutual TLS between pairs of services that need to talk to each other.

Default runs as root $ docker run -v /bin:/host/bin -it --rm alpine sh / $ whoami root / $ id uid=0(root) gid=0(root) / $ # WREAK HAVOC TIME! Please don’t do this / $ rm /host/bin/sh # Again, please don’t do this

root in container == root outside container We don’t want this to be the case! How can we change this? 0 (root) 1000 (user) 65534 (nobody)

Step in the right direction: run as a user # Use the --user flag with UID:GID argument $ docker run -v /bin:/host/bin --user 10000:10000 -it --rm alpine sh / $ whoami whoami: unknown uid 10000 / $ id uid=10000 gid=10000 / $ rm /host/bin/sh rm: can’t remove ‘sh’: Permission denied

But I still want “root” inside container Perhaps we need to run a command that needs to look like it’s root in the container, but we don’t want to give it true root access to the underlying host 0 (root) 1000 (user) 65534 (nobody) 10000 (docker)

$ docker daemon --userns-remap [uid[:gid]] Enable user namespaces 0 (root) 1000 (user) 65534 (nobody) 10000 (docker) “0” “root”

$ docker daemon --userns-remap [uid[:gid]] Will need to re-pull images and re-create volumes due to container resource and image layer permissions Leave this feature on in production; switching back and forth should only be done in development Enable user namespaces - common pitfalls 0 (root) 1000 (user) 65534 (nobody) 10000 (docker) “0” “root”

Hands-On Exercise && break github.com/riyazdf/dockercon-workshop - userns directory

Security Goals Image Provenance and Trust ● Provenance: who made this image? ○ Verify the publisher of the image ● Trust: have the contents of this image been tampered with? ○ Verify the integrity of the image

Pulling by tag $ docker pull alpine:latest Name resolution takes place in registry to find content-address of image $ docker pull alpine Using default tag: latest Notice that the tag defaults to latest if no tags are given!

Pulling by digest $ docker pull alpine@sha256:ea0d1389812... No name resolution! Security best practice: pulling by digest to enforce consistent and “immutable” pulls because of content-addressability

Content Trust $ export DOCKER_CONTENT_TRUST=1 $ docker pull alpine:latest Pull (1 of 1): alpine:latest@sha256:ea0d1389... Benefits of pull by digest with ease of pull by tag

docker pull alpine:latest Without trust enabled: forward command through Docker Engine to registry to resolve tag to digest

Content Trust (on push) $ export DOCKER_CONTENT_TRUST=1 $ docker tag alpine:latest <user>/alpine:trust $ docker push <user>/alpine:trust Looks the same as a regular push by tag!

Content Trust (it’s more than gpg) The push refers to a repository [<user>/alpine] 77f08abee8bf: Pushed trust: digest: sha256:d5de850d728... size: 1355 Signing and pushing trust metadata Enter passphrase for root key with ID e83f424: Enter passphrase for new repository key with ID f903fc9 (docker.io/<user>/alpine): Repeat passphrase for new repository key with ID f903fc9 (docker.io/<user>/alpine): Finished initializing "docker.io/<user>/alpine" Successfully signed "docker.io/<user>/alpine":trust

Content Trust (it’s more than gpg) $ cat ~/.docker/trust/tuf/docker.io/alpine/metadata/timestamp.json | jq

Docker Content Trust / Notary Threat Model ● Key compromise? ○ We can recover! ● Replay attacks? ○ Not with our freshness guarantees! ● Untrusted registry? ○ No problem! DCT/Notary do not root any trust in the underlying content store or transport ○ Use signed TUF metadata to retrieve trusted hashes of content ○ Don’t even need to trust Notary server after first pull - local metadata pins trust, tagging keys are kept client-side for signing

Docker Pull Only pull trusted images Use official images when possible!

Docker Security Scanning (Nautilus)

Docker Security Scanning (Nautilus) https://hub.docker.com/r/library/alpine/tags/ - All official images on hub are scanned for vulnerabilities, lobby upstream for fixes! - Can view scan results after logging into Docker Hub

Docker Security Scanning (Nautilus) - Checks against CVE database for declared layers - Also performs binary scan to pick up on statically linked binaries - Official repos have been scanned since Nov 2015, are rescanned often

Hands-On Exercise github.com/riyazdf/dockercon-workshop - trust directory

Root vs Not Root Capabilities breakdown root permissions into groups that can be individually allowed or blocked - Often don’t want or need all root permissions - Can reduce attack surface by reducing capabilities

Docker Default Capabilities In whitelist: "CAP_CHOWN", "CAP_DAC_OVERRIDE", "CAP_FSETID", "CAP_FOWNER", "CAP_MKNOD", "CAP_NET_RAW", "CAP_SETGID", "CAP_SETUID", "CAP_SETFCAP", "CAP_SETPCAP", "CAP_NET_BIND_SERVICE", "CAP_SYS_CHROOT", "CAP_KILL", "CAP_AUDIT_WRITE", Not in whitelist: "CAP_AUDIT_CONTROL", "CAP_AUDIT_READ", "CAP_BLOCK_SUSPEND", "CAP_DAC_READ_SEARCH", "CAP_IPC_LOCK", "CAP_ICP_OWNER", "CAP_LEASE", "CAP_LINUX_IMMUTABLE", "CAP_MAC_ADMIN", "CAP_MAC_OVERRIDE", "CAP_NET_ADMIN", "CAP_NET_BROADCAST", "CAP_SYS_ADMIN", "CAP_SYS_BOOT", "CAP_SYS_MODULE", "CAP_SYS_NICE", "CAP_SYS_PACCT", "CAP_SYS_PTRACE", "CAP_SYS_RAWIO", "CAP_SYS_RESOURCE", "CAP_SYS_TIME", "CAP_SYS_TTY_CONFIG", "CAP_SYSLOG", “CAP_WAKE_ALARM",

How do we add/remove capabilities? docker run --cap-add docker run --cap-drop docker run --cap-drop ALL --cap-add $CAP

Configure capabilities in compose cap_add: - CAP_NET_BROADCAST - CAP_NET_RAW cap_drop: - ALL

What to watch out for - Read the fine print for each capability! - man capabilities - i.e. removing CAP_KILL only requires permissions checks and enabling bypasses permissions checks. It doesn’t generically enable/disable the ability to kill - CAP_SYS_ADMIN is nearly root...

What to watch out for $ man capabilities ... CAP_SYS_ADMIN * Perform a range of system administration operations including: quotactl(2), mount(2), umount(2), swapon(2), setdomainname(2); * perform privileged syslog(2) operations (since Linux 2.6.37, CAP_SYSLOG should be used to permit such operations); * perform VM86_REQUEST_IRQ vm86(2) command; * perform IPC_SET and IPC_RMID operations on arbitrary System V IPC objects; * override RLIMIT_NPROC resource limit; * perform operations on trusted and security Extended Attributes (see xattr(7)); * use lookup_dcookie(2); * use ioprio_set(2) to assign IOPRIO_CLASS_RT and (before Linux 2.6.25) IOPRIO_CLASS_IDLE I/O scheduling classes; * forge PID when passing socket credentials via UNIX domain sockets; * exceed /proc/sys/fs/file-max, the system-wide limit on the number of open files, in system calls that open files (e.g., accept(2), execve(2), open(2), pipe(2)); * employ CLONE_* flags that create new namespaces with clone(2) and unshare(2) (but, since Linux 3.8, creating user namespaces does not require any capability); * call perf_event_open(2); * access privileged perf event information; * call setns(2) (requires CAP_SYS_ADMIN in the target

Capabilities and Docker ● No extended attributes in images -> no capabilities elevation normally possible ● Use docker to reduce capabilities ● Docker can’t grant capabilities to non-root users due to some limitations in older kernel versions

Capabilities and Docker Your options from worst to best: 1. Manual management within the container: docker run --cap-add ALL 2. Restricted capabilities (still root): docker run --cap-drop ALL --cap-add ABC 3. No capabilities: docker run --user

What to watch out for $ docker run --privileged … gives all capabilities to the container, also lifts limitations from device cgroup

More information github.com/riyazdf/dockercon-workshop - capabilities directory

Original Seccomp On-off feature that disabled all system calls except: exit() read() write() sigreturn()

Seccomp-BPF Extension Allows us to configure what system calls are allowed/blocked Uses Berkeley Packet Filters (BPF) Allows examining system calls in detail before making a decision

Is it enabled? In the kernel: $ grep SECCOMP /boot/config-$(uname -r) # or zgrep SECCOMP /proc/config.gz CONFIG_SECCOMP=y CONFIG_SECCOMP_FILTER=y In docker: $ docker run --rm alpine grep Seccomp /proc/self/status In docker 1.12: $ docker info

Default Whitelist Lots of system calls, what’s excluded: acct add_key adjtimex bpf clock_adjtime clock_settime clone create_module delete_module finit_module get_kernel_syms get_mempolicy init_module ioperm iopl kcmp kexec_file_load kexec_load keyctl lookup_dcookie mbind mount move_pages name_to_handle_at nfsservctl open_by_handle_at perf_event_open personality pivot_root process_vm_readv process_vm_writev ptrace query_module quotactl reboot request_key set_mempolicy setns settimeofday stime swapon swapoff sysfs _sysctl umount umount2 unshare uselib userfaultfd ustat vm86 vm86old

The strace tool $ strace -c -f -S name ls 2>&1 1>/dev/null | tail -n +3 | head -n -2 | awk '{print $(NF)}' access arch_prctl brk close execve fstat getdents ioctl mmap mprotect munmap open read write

Docker seccomp profile DSL Seccomp policy example: { "defaultAction": "SCMP_ACT_ERRNO", "architectures": [ "SCMP_ARCH_X86_64", "SCMP_ARCH_X86", "SCMP_ARCH_X32" ], "syscalls": [ { "name": "accept", "action": "SCMP_ACT_ALLOW", "args": [] }, ... ] } Possible actions: SCMP_ACT_KILL SCMP_ACT_TRAP SCMP_ACT_ERRNO SCMP_ACT_TRACE SCMP_ACT_ALLOW

Docker seccomp profile DSL More complex filters: "args": [ { "index": 0, "op": "SCMP_CMP_MASKED_EQ", "value": 2080505856, "valueTwo": 0 } ] 2080505856 == 0x7C020000

Seccomp and the no-new-privileges option Seccomp policies have to be applied before executing your container and be less specific unless you use: --security-opt no-new-privileges In this case you need to allow only futex stat execve. This flag also disables setuid binaries: $ sudo ls sudo: effective uid is not 0, is /usr/bin/sudo on a file system with the 'nosuid' option set or an NFS file system without root privileges?

More information github.com/riyazdf/dockercon-workshop - seccomp directory $ docker run --rm -it --security-opt seccomp=default-no-chmod.json alpine chmod 777 / chmod: /: Operation not permitted

What is a LSM? A plugin to the linux kernel that allows us to set policies to restrict what a process can do. Mandatory Access Control: instead of using user-defined permissions to specify access, the underlying system describes permissions itself with labels

What is a LSM? Under the hood: each LSM implements a kernel interface that hooks into user-level syscalls about to access an important kernel object (inodes, task control blocks, etc.), either allowing them to pass through or denying them outright depending on the application profile

Available LSMs AppArmor SELinux Smack Tomoyo

Deep Dive - AppArmor File Access Management AppArmor uses globbing and deny syntax to express filepath restrictions ● deny /sys/* rwklx Deny read/write/lock/link/execute on files in /sys/ ● deny /sys/** rwklx Deny on files in /sys/ and subdirectories

Deep Dive - AppArmor Networking Management Like firewall rules: ● Can completely disable networking: deny network ● Can deny certain permissions: deny network bind, inet ● Can specify specific IP/ports : network tcp src 192.168.1.1:80 dst 170.1.1.0:80

Deep Dive - AppArmor Capability Management AppArmor can also deny capabilities with a simple syntax: ● deny capability chown, ● deny capability dac_override

Deep Dive - AppArmor Composability C-style include statements ● include <abstractions/base> - built-in bundle of files ● include "/etc/apparmor.d/include/foo" - absolute path from file ● include "../relative_path/bar" - relative path from file

Deep Dive - AppArmor Tools for debugging and generating profiles (on Ubuntu): $ sudo apt install apparmor-utils $ aa-complain <PATH_TO_PROFILE> # Watch AppArmor block things! $ aa-genprof <PATH_TO_BINARY> # Interactive profile generation!

Do I still need Seccomp and Cap-drop? Why not? Docker sets a profile for each setting by default ● Some overlap but each feature still adds unique functionality ● Defense-in-depth

Common mistake: disabling profiles SELinux: setenforce 0 (on daemon) http://stopdisablingselinux.com/ AppArmor: --security-opt apparmor:unconfined (on docker run) This one’s a little harder to do “by accident” docker run --privileged

Hands-On Exercise github.com/riyazdf/dockercon-workshop - apparmor directory

Docker Bench https://dockerbench.com - Open-source tool for running automated tests - inspired by the CIS Docker 1.11 benchmark - Runs against containers currently running on same host - Checks for AppArmor, read-only volumes, etc...

View from 10,000 feet: Docker Security Checklist Build: - Use minimal images (alpine) - Use official images - Using images pulled by content trust (fresh, pulled by digest from authors you trust) Ship: - Push to your consumers with content trust - View results from Docker Security Scanning Run: - Mutual TLS between client/engine - Read-only volumes and containers - User namespaces in the daemon - Limit resources with cgroups - Use the default apparmor/seccomp/capabilities, or your own tested profiles (not --privileged!)

Thank you! - Please take the exercises home! We’ll accept issues and pull requests :) - Learn more at https://docs.docker.com/engine/security/security/

Advanced Topics “Extra for Experts”

Running your own Notary Deploy a notary $ git clone https://github.com/docker/notary.git $ cd notary $ docker-compose up

Notary Delegations (admin)$ notary key rotate <GUN> snapshot -r (user)$ < generates private key and x509 cert, gives user.crt to admin > (admin)$ notary delegation add <GUN> targets/user user.crt --all-paths (admin)$ notary publish <GUN> Docker engine >= 1.11 will sign with delegation keys if it detects them

Docker Security workshop slides

Docker Security workshop slides

In this document