Originally published at https://www.prabhujayakumar.dev
In this blog, we will look at a kubernetes feature which is intended to improve security for applications running in a cluster
In a kubernetes cluster, we can run many applications with multiple replicas for each application. By default, any pods can talk to any other pods running in the same cluster.
But it is not recommended to allow such feature to be in place. In case an intruder get access to a pod, then he/she can access all pods from inside that compromised pod. So we need a firewall for applications using which it can decide if the traffic(both ingress and egress) should be allowed or denied inside the cluster.
There comes the saviour, Network Policy
that helps to create a firewall for applications running in kubernetes cluster.
Let's understand the need for such firewall and how Network Policy helps for the same with some examples.
Guestbook
Consider an application Guestbook
having 3 different components as follows:
- guestbook-ui (frontend)
- guestbook-api (backend)
- guestbook-db (db)
Expected communication between these components are like ui communicates to api and api communicates to db.
But when all these components of guestbook application are running in a kubernetes cluster, technically ui component can communicate to db by default.
Use Network Policy to setup firewall for applications in cluster
As network policy is a feature that should be implemented by the network plugin, ensure that your network plugin supports NetworkPolicy resource. Creating NetworkPolicy resource without such plugin will have no effect. Calico, Cilium, Kube-router, Romana and Weave Net are some of the network plugins that support network policy
As a first step, disable the default behaviour of allowing all communications between all pods running in a kubernetes cluster by creating a network policy as follows in all namespaces.
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
name: deny-all
spec:
podSelector: {}
ingress: {}
Now, none of the pods can communicate to any other pods running in the kubernetes cluster. Now based on requirements, allow the ingress/egress traffic for pods in the cluster.
For the guestbook application, to allow ingress traffic to guestbook-api
pods but only from guestbook-ui
pods, create a network policy as follows.
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
name: allow-ui-to-api
spec:
podSelector:
matchLabels:
app: guestbook-api
tier: backend
ingress:
- from:
podSelector:
matchLabels:
app: guestbook-ui
tier: frontend
And to allow ingress traffic to guestbook-db
pods but only from guestbook-api
pods, create a network policy as follows.
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
name: allow-api-to-db
spec:
podSelector:
matchLabels:
app: guestbook-db
tier: db
ingress:
- from:
podSelector:
matchLabels:
app: guestbook-api
tier: backend
Final setup with these network policies allows only the valid communications between the pods
More about network policy
NamespaceSelector
In some scenarios, we have to allow ingress from any pods in a namespace. For example, all applications pods should allow ingress from all pods in monitoring
namespace. Namespace selector can be used to achieve this setup easily as follows.
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
name: allow-monitoring-namespace
spec:
podSelector: {} ## applies to all pods in the namespace
ingress:
- from:
namespaceSelector:
matchLabels:
team: monitoring ## labels of the monitoring namespace
IP Block
Not just labels, network policy also allows configuration using IP blocks. Following network policy allows ingress from 10.72.X.X but blocks traffic from 10.72.10.X
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
name: allow-using-ip
spec:
podSelector: {}
ingress:
- from:
ipBlock:
cidr: 10.72.0.0/16
except: 10.72.10.0/8
Conclusion
It is good to have security at all levels in k8s cluster. Adding a firewall for the application pods in the kubernetes cluster increases level of security by reducing the attack surface and hence is highly recommended.
Top comments (0)