Friday, December 7, 2018

Kubernetes Privilege Escalation (CVE-2018-1002105)


Kubernetes is an open source production grade container orchestration system for deploying and managing docker/container applications. There are managed kubernetes orchestration service providers like Amazon Elastic Container Service for Kubernetes (EKS), Azure Kubernetes Service (AKS) etc.


Kubernetes cluster users can perform management tasks using kubectl binary which talks to API Server. Example kubectl commands

# display pod resource
kubectl get pods -n my_namespace

# Execute a command in a container
kubectl -n my_namespace exec -it pods_name -- sh

# Listen on ports 5000 and 6000 locally, forwarding data to/from ports 5000 and 6000 in the pod
kubectl -n my_namespace port-forward pod/mypod 5000 6000

# Get output from ruby-container from pod my-pod-pd

kubectl attach my-pod-pd -c ruby-container

kubectl execution flow (source:


kubelet, kube-proxy run's on each compute node (VM, Worker, EC2 Instance etc), kubelet listens on TCP port 10250 and 10255 (with no authentication/authorization). API Server acts as Reverse Proxy to kubelet and API Aggregation. API Server connects to the kubelet to fulfill commands like exec, port=forward and opens a websocket connection which connects stdin, stdout, or stderr to user’s original call [01].

API Aggregation

Installing or writing additional API's into Kubernetes API Server i.e. extending core API Server


Vulnerability is in Kubernetes API Server, crafted request can execute arbitrary commands on the backend servers (pods) through the same channel client established to backend through API Server [02]

Check nodes Kubernetes version
$ kubectl get nodes -o wide
pd-worker-01 Ready node 13d v1.12.3 Container Linux by CoreOS 1745.7.0 (Rhyolite) 4.14.48-coreos-r2 docker://18.3.1
pd-worker-02 Ready node 13d v1.12.3 Container Linux by CoreOS 1745.7.0 (Rhyolite) 4.14.48-coreos-r2 docker://18.3.1
pd-worker-03 Ready node 13d v1.12.3 Container Linux by CoreOS 1745.7.0 (Rhyolite) 4.14.48-coreos-r2 docker://18.3.1

Vulnerable API Servers

If API server response looks as bellow and using vulnerable API versions of Kubernetes the you are vulnerable using anonymous-user escalation, patch Kubernetes immediately.
HTTP response error code 403 indicates Forbidden i.e. related to Authorization implies we successfully passed through Authentication phase.
{ "kind": "Status", "apiVersion": "v1", "metadata": { }, "status": "Failure", "message": "forbidden: User \"system:anonymous\" cannot get path \"/api/v1/\"", "reason": "Forbidden", "details": { }, "code": 403 }

anonymous user

By default, requests to the kubelet’s HTTPS endpoint that are not rejected by other configured authentication methods are treated as anonymous requests, and given a username of system:anonymous and a group of system:unauthenticated.


There are three levels of escalation mitigations

1. anonymous user -> aggregated API server

API Server admission-controller parameter anonymous-auth is set to fault
$ kubectl get po kube-apiserver-01 -n prod -o yaml | grep -i "anonymous-auth" - --anonymous-auth=false 
$ kubectl get po kube-apiserver-01 -n stage -o yaml | grep -i "anonymous-auth" - --anonymous-auth=false

2. authenticated user -> aggregated API server

Suspend aggregated API servers usage

3. authorized pod exec/attach/portforward -> kubelet API

Remove pod exec/attach/portforward permissions for users



Saturday, January 20, 2018

AWS VPC Flow Logs grok Pattern

Amazon Web Services(AWS) can generate VPC flow logs, format below
2 123456789010 eni-abc123de 49761 3389 6 20 4249 1418530010 1418530070 REJECT OK

For more information on flow logs and grok filter plugin refer below links

grok patterns can be tested using below links

%{NONNEGINT:version} %{NONNEGINT:accountid} %{NOTSPACE:interface-id} %{NOTSPACE:srcaddr} %{NOTSPACE:dstaddr} %{NONNEGINT:srcport} %{NONNEGINT:dstport} %{NONNEGINT:protocol} %{NONNEGINT:packets} %{NONNEGINT:bytes} %{NONNEGINT:starttime} %{NONNEGINT:endtime} %{NOTSPACE:action} %{NOTSPACE:log-status}

Test using grokdebugger

Test using grokconstructor

You can also consider INT instead of NONNEGINT

Found few patterns by googling which looked like below, were not working on grokconstructor website.
%{NUMBER:version} %{NUMBER:account-id} %{NOTSPACE:interface-id} %{NOTSPACE:srcaddr} %{NOTSPACE:dstaddr} %{NOTSPACE:srcport:int} %{NOTSPACE:dstport:int} %{NOTSPACE:protocol:int} %{NOTSPACE:packets:int} %{NOTSPACE:bytes:int} %{NUMBER:start:int} %{NUMBER:end:int} %{NOTSPACE:action} %{NOTSPACE:log-status}

Tested on grokdebugger

Tested on grokconstructor

We can use the extracted variables from grok filter plugin in Kibana search or enhance data using logstash filter plugins geoip, dns, date etc.

Working in or using Python virtualenv

Install Python virtualenv on Ubuntu using below command
apt-get -y install python-virtualenv

Create virtualenv
$virtualenv test_env1
New python executable in test_env1/bin/python

$. test_env1/bin/activate
$source test_env1/bin/activate

Exit virtualenv

Switch between virtualenv’s
$workon test_env2

List all available virtualenv’s

virtualenvwrapper comes with few handy commands

$pip install virtualenvwrapper

virtualenvwrapper supports extra commands like