Network Policy

Purpose

The purpose of this document is to present how Network Policy is supported by Kuryr-Kubernetes.

Overview

Kubernetes supports a Network Policy object to express ingress and egress rules for pods. Network Policy reacts on labels to qualify multiple pods, and defines rules based on differents labeling and/or CIDRs. When combined with a networking plugin, those policy objetcs are enforced and respected.

Proposed Solution

Kuryr-Kubernetes relies on Neutron security groups and security group rules to enforce a Network Policy object, more specifically one security group per policy with possibly multiple rules. Each object has a namespace scoped Network Policy CRD that stores all OpenStack related resources on the Kubernetes side, avoiding many calls to Neutron and helping to differentiate between the current Kubernetes status of the Network Policy and the last one Kuryr-Kubernetes enforced.

The network policy CRD has the following format:

apiVersion: openstack.org/v1
kind: KuryrNetPolicy
metadata:
  ...
spec:
  egressSgRules:
  - security_group_rule:
    ...
  ingressSgRules:
  - security_group_rule:
    ...
  networkpolicy_spec:
    ...
  podSelector:
    ...
  securityGroupId: ...
  securityGroupName: ...

A new handler has been added to react to Network Policy events, and the existing ones, for instance service/pod handlers, have been modified to account for the side effects/actions of when a Network Policy is being enforced.

Note

Kuryr supports a network policy that contains:

  • Ingress and Egress rules

  • namespace selector and pod selector, defined with match labels or match expressions, mix of namespace and pod selector, ip block

  • named port

New handlers and drivers

The Network Policy handler

This handler is responsible for triggering the Network Policy Spec processing, and the creation or removal of security group with appropriate security group rules. It also, applies the security group to the pods and services affected by the policy.

The Pod Label handler

This new handler is responsible for triggering the update of a security group rule upon pod labels changes, and its enforcement on the pod port and service.

The Network Policy driver

Is the main driver. It ensures a Network Policy by processing the Spec and creating or updating the Security group with appropriate security group rules.

The Network Policy Security Group driver

It is responsible for creating, deleting, or updating security group rules for pods, namespaces or services based on different Network Policies.

Modified handlers and drivers

The VIF handler

As network policy rules can be defined based on pod labels, this handler has been enhanced to trigger a security group rule creation or deletion, depending on the type of pod event, if the pod is affected by the network policy and if a new security group rule is needed. Also, it triggers the translation of the pod rules to the affected service.

The Namespace handler

Just as the pods labels, namespaces labels can also define a rule in a Network Policy. To account for this, the namespace handler has been incremented to trigger the creation, deletion or update of a security group rule, in case the namespace affects a Network Policy rule, and the translation of the rule to the affected service.

The Namespace Subnet driver

In case of a namespace event and a Network Policy enforcement based on the namespace, this driver creates a subnet to this namespace, and restrict the number of security group rules for the Network Policy to just one with the subnet CIDR, instead of one for each pod in the namespace.

The LBaaS driver

To restrict the incoming traffic to the backend pods, the LBaaS driver has been enhanced to translate pods rules to the listener port, and react to Service ports updates. E.g., when the target port is not allowed by the policy enforced in the pod, the rule should not be added.

The VIF Pool driver

The VIF Pool driver is responsible for updating the Security group applied to the pods ports. It has been modified to embrace the fact that with Network Policies pods’ ports changes their security group while being used, meaning the original pool does not fit them anymore, resulting in useless pools and ports reapplying the original security group. To avoid it, the security group id is removed from the pool merging all pools with same network, project and host id. Thus if there is no ports on the pool with the needed security group id(s), one of the existing ports in the pool is updated to match the requested sg Id.

Use cases examples

This section describes some scenarios with a Network Policy being enforced, what Kuryr componenets gets triggered and what resources are created.

Deny all incoming traffic

By default, Kubernetes clusters do not restrict traffic. Only once a network policy is enforced to a namespace, all traffic not explicitly allowed in the policy becomes denied. As specified in the following policy:

apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: default-deny
spec:
  podSelector: {}
  policyTypes:
  - Ingress

The following CRD is the translation of policy rules to security group rules. No ingress rule was created, which means traffic is blocked, and since there is no restriction for egress traffic, it is allowed to everywhere. Note that the same happens when no policyType is defined, since all policies are assumed to assumed to affect Ingress.

apiVersion: openstack.org/v1
kind: KuryrNetPolicy
metadata:
  name: np-default-deny
  namespace: default
  ...
spec:
  egressSgRules:
  - security_group_rule:
      description: Kuryr-Kubernetes NetPolicy SG rule
      direction: egress
      ethertype: IPv4
      id: 60a0d59c-2102-43e0-b025-75c98b7d9315
      security_group_id: 20d9b623-f1e0-449d-95c1-01624cb3e315
  ingressSgRules: []
  networkpolicy_spec:
    ...
  podSelector:
    ...
  securityGroupId: 20d9b623-f1e0-449d-95c1-01624cb3e315
  securityGroupName: sg-default-deny

Allow traffic from pod

The following Network Policy specification has a single rule allowing traffic on a single port from the group of pods that have the label role=monitoring.

apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: allow-monitoring-via-pod-selector
spec:
  podSelector:
    matchLabels:
      app: server
  policyTypes:
  - Ingress
  ingress:
  - from:
    - podSelector:
        matchLabels:
          role: monitoring
    ports:
    - protocol: TCP
      port: 8080

Create the following pod with label role=monitoring:

$ kubectl run  monitor --image=busybox --restart=Never --labels=role=monitoring

The generated CRD contains an ingress rule allowing traffic on port 8080 from the created pod, and an egress rule allowing traffic to everywhere, since no restriction was enforced.

apiVersion: openstack.org/v1
kind: KuryrNetPolicy
metadata:
  name: np-allow-monitoring-via-pod-selector
  namespace: default
  ...
spec:
  egressSgRules:
  - security_group_rule:
      description: Kuryr-Kubernetes NetPolicy SG rule
      direction: egress
      ethertype: IPv4
      id: 203a14fe-1059-4eff-93ed-a42bd957145d
      security_group_id: 7f0ef8c2-4846-4d8c-952f-94a9098fff17
  ingressSgRules:
  - namespace: default
    security_group_rule:
      description: Kuryr-Kubernetes NetPolicy SG rule
      direction: ingress
      ethertype: IPv4
      id: 7987c382-f2a9-47f7-b6e8-1a3a1bcb7d95
      port_range_max: 8080
      port_range_min: 8080
      protocol: tcp
      remote_ip_prefix: 10.0.1.143
      security_group_id: 7f0ef8c2-4846-4d8c-952f-94a9098fff17
  networkpolicy_spec:
    ...
  podSelector:
    ...
  securityGroupId: 7f0ef8c2-4846-4d8c-952f-94a9098fff17
  securityGroupName: sg-allow-monitoring-via-pod-selector

Allow traffic from namespace

The following network policy only allows allowing ingress traffic from namespace with the label purpose=test:

apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: allow-test-via-ns-selector
spec:
  podSelector:
    matchLabels:
      app: server
  policyTypes:
  - Ingress
  ingress:
  - from:
    - namespaceSelector:
        matchLabels:
          purpose: test
    ports:
    - protocol: TCP
      port: 8080

Create a namespace and label it with purpose=test:

$ kubectl create namespace dev
$ kubectl label namespace dev purpose=test

The resulting CRD has an ingress rule allowing traffic from the namespace CIDR on the specified port, and an egress rule allowing traffic to everywhere.

apiVersion: openstack.org/v1
kind: KuryrNetPolicy
  name: np-allow-test-via-ns-selector
  namespace: default
  ...
spec:
  egressSgRules:
  - security_group_rule:
      description: Kuryr-Kubernetes NetPolicy SG rule
      direction: egress
      ethertype: IPv4
      id: 8c21bf42-c8b9-4628-b0a1-bd0dbb192e6b
      security_group_id: c480327c-2db4-4eb6-af1e-eeb0ce9b46c9
  ingressSgRules:
  - namespace: dev
    security_group_rule:
      description: Kuryr-Kubernetes NetPolicy SG rule
      direction: ingress
      ethertype: IPv4
      id: 2a33b802-56ad-430a-801d-690f653198ef
      port_range_max: 8080
      port_range_min: 8080
      protocol: tcp
      remote_ip_prefix: 10.0.1.192/26
      security_group_id: c480327c-2db4-4eb6-af1e-eeb0ce9b46c9
  networkpolicy_spec:
    ...
  podSelector:
    ...
  securityGroupId: c480327c-2db4-4eb6-af1e-eeb0ce9b46c9
  securityGroupName: sg-allow-test-via-ns-selector

Note

The Service security groups need to be rechecked when a network policy that affects ingress traffic is created, and also everytime a pod or namespace is created.

Create network policy flow

Network Policy creation flow

Create pod flow

The following diagram only covers the implementation part that affects network policy.

Pod creation flow

Network policy rule definition

NamespaceSelector

podSelector

Expected result

namespaceSelector: ns1

podSelector: pod1

Allow traffic from pod1 at ns1

namespaceSelector: ns1

podSelector: {}

Allow traffic from all pods at ns1

namespaceSelector: ns1

none

Allow traffic from all pods at ns1

namespaceSelector: {}

podSelector: pod1

Allow traffic from pod1 from all namespaces

namespaceSelector: {}

podSelector: {}

Allow traffic from all namespaces

namespaceSelector: {}

none

Allow traffic from all namespaces

none

podSelector: pod1

Allow traffic from pod1 from NP namespace

none

podSelector: {}

Allow traffic from all pods from NP namespace

Rules definition

Expected result

No FROM (or from: [])

Allow traffic from all pods from all namespaces

Ingress: {}

Allow traffic from all namespaces

ingress: []

Deny all traffic

No ingress

Blocks all traffic

Policy types definition

PolicyType

Spec Ingress/Egress

Ingress generated rules

Egress generated rules

none

none

BLOCK

ALLOW

none

ingress

Specific rules

ALLOW

none

egress

Block

Specific rules

none

ingress, egress

Specific rules

Specific rules

ingress

none

Block

ALLOW

ingress

ingress

Specific rules

ALLOW

egress

none

ALLOW

BLOCK

egress

egress

ALLOW

Specific rules

Ingress, egress

none

BLOCK

BLOCK

Ingress, egress

ingress

Specific rules

BLOCK

Ingress, egress

egress

BLOCK

Specific rules

Ingress, egress

ingress,egress

Specific rules

Specific rules