Security

This section contains information about security for CloudNativePG, that are analyzed at 3 different layers: Code, Container and Cluster.

Code

CloudNativePG's source code undergoes systematic static analysis, including checks for security vulnerabilities, using the popular open-source linter for Go, GolangCI-Lint, directly integrated into the CI/CD pipeline. GolangCI-Lint can run multiple linters on the same source code.

The following tools are used to identify security issues:

• Golang Security Checker (gosec): A linter that scans the abstract syntax tree of the source code against a set of rules designed to detect known vulnerabilities, threats, and weaknesses, such as hard-coded credentials, integer overflows, and SQL injections. GolangCI-Lint runs gosec as part of its suite.

• govulncheck: This tool runs in the CI/CD pipeline and reports known vulnerabilities affecting Go code or the compiler. If the operator is built with a version of the Go compiler containing a known vulnerability, govulncheck will detect it.

• CodeQL: Provided by GitHub, this tool scans for security issues and blocks any pull request with detected vulnerabilities. CodeQL is configured to review only Go code, excluding other languages in the repository such as Python or Bash.

• Snyk: Conducts nightly code scans in a scheduled job and generates weekly reports highlighting any new findings related to code security and licensing issues.

The CloudNativePG repository has the "Private vulnerability reporting" option enabled in the Security section. This feature allows users to safely report security issues that require careful handling before being publicly disclosed. If you discover any security bug, please use this medium to report it.

Container

Every container image in CloudNativePG is automatically built via CI/CD pipelines following every commit. These images include not only the operator's image but also the operands' images, specifically for every supported PostgreSQL version. During the CI/CD process, images undergo scanning with the following tools:

• Dockle: Ensures best practices in the container build process.
• Snyk: Detects security issues within the container and reports findings via the GitHub interface.

Guidelines and Frameworks for Container Security

The following guidelines and frameworks have been considered for ensuring container-level security:

• "Container Image Creation and Deployment Guide": Developed by the Defense Information Systems Agency (DISA) of the United States Department of Defense (DoD).
• "CIS Benchmark for Docker": Developed by the Center for Internet Security (CIS).

Cluster

Security at the cluster level takes into account all Kubernetes components that form both the control plane and the nodes, as well as the applications that run in the cluster (PostgreSQL included).

Role Based Access Control (RBAC)

The operator interacts with the Kubernetes API server using a dedicated service account named cnpg-manager. This service account is typically installed in the operator namespace, commonly cnpg-system. However, the namespace may vary based on the deployment method (see the subsection below).

In the same namespace, there is a binding between the cnpg-manager service account and a role. The specific name and type of this role (either Role or ClusterRole) also depend on the deployment method. This role defines the necessary permissions required by the operator to function correctly. To learn more about these roles, you can use the kubectl describe clusterrole or kubectl describe role commands, depending on the deployment method.

Below we provide some examples and, most importantly, the reasons why CloudNativePG requires full or partial management of standard Kubernetes namespaced or non-namespaced resources.

configmaps

The operator needs to create and manage default config maps for the Prometheus exporter monitoring metrics.

deployments

The operator needs to manage a PgBouncer connection pooler using a standard Kubernetes Deployment resource.

jobs

The operator needs to handle jobs to manage different Cluster's phases.

persistentvolumeclaims

The volume where the PGDATA resides is the central element of a PostgreSQL Cluster resource; the operator needs to interact with the selected storage class to dynamically provision the requested volumes, based on the defined scheduling policies.

pods

The operator needs to manage Cluster's instances.

secrets

Unless you provide certificates and passwords to your Cluster objects, the operator adopts the "convention over configuration" paradigm by self-provisioning random generated passwords and TLS certificates, and by storing them in secrets.

serviceaccounts

The operator needs to create a service account that enables the instance manager (which is the PID 1 process of the container that controls the PostgreSQL server) to safely communicate with the Kubernetes API server to coordinate actions and continuously provide a reliable status of the Cluster.

services

The operator needs to control network access to the PostgreSQL cluster (or the connection pooler) from applications, and properly manage failover/switchover operations in an automated way (by assigning, for example, the correct end-point of a service to the proper primary PostgreSQL instance).

validatingwebhookconfigurations and mutatingwebhookconfigurations

The operator injects its self-signed webhook CA into both webhook configurations, which are needed to validate and mutate all the resources it manages. For more details, please see the Kubernetes documentation.

volumesnapshots

The operator needs to generate VolumeSnapshots objects in order to take backups of a PostgreSQL server. VolumeSnapshots are read too in order to validate them before starting the restore process.

nodes

The operator needs to get the labels for Affinity and AntiAffinity so it can decide in which nodes a pod can be scheduled. This is useful, for example, to prevent the replicas from being scheduled in the same node - especially important if nodes are in different availability zones. This permission is also used to determine whether a node is scheduled, preventing the creation of pods on unscheduled nodes, or triggering a switchover if the primary lives in an unscheduled node.

Deployments and ClusterRole Resources

As mentioned above, each deployment method may have variations in the namespace location of the service account, as well as the names and types of role bindings and respective roles.

Via Kubernetes Manifest

When installing CloudNativePG using the Kubernetes manifest, permissions are set to ClusterRoleBinding by default. You can inspect the permissions required by the operator by running:

kubectl describe clusterrole cnpg-manager

Via OLM

From a security perspective, the Operator Lifecycle Manager (OLM) provides a more flexible deployment method. It allows you to configure the operator to watch either all namespaces or specific namespaces, enabling more granular permission management.

OLM allows you to deploy the operator in its own namespace and configure it to watch specific namespaces used for CloudNativePG clusters. This setup helps to contain permissions and restrict access more effectively.

Why Are ClusterRole Permissions Needed?

The operator currently requires ClusterRole permissions to read nodes and ClusterImageCatalog objects. All other permissions can be namespace-scoped (i.e., Role) or cluster-wide (i.e., ClusterRole).

Even with these permissions, if someone gains access to the ServiceAccount, they will only have get, list, and watch permissions, which are limited to viewing resources. However, if an unauthorized user gains access to the ServiceAccount, it indicates a more significant security issue.

Therefore, it's crucial to prevent users from accessing the operator's ServiceAccount and any other ServiceAccount with elevated permissions.

Calls to the API server made by the instance manager

The instance manager, which is the entry point of the operand container, needs to make some calls to the Kubernetes API server to ensure that the status of some resources is correctly updated and to access the config maps and secrets that are associated with that Postgres cluster. Such calls are performed through a dedicated ServiceAccount created by the operator that shares the same PostgreSQL Cluster resource name.

For transparency, the permissions associated with the service account are defined in the roles.go file. For example, to retrieve the permissions of a generic mypg cluster in the myns namespace, you can type the following command:

kubectl get role -n myns mypg -o yaml

Then verify that the role is bound to the service account:

kubectl get rolebinding -n myns mypg -o yaml

Below we provide a quick summary of the permissions associated with the service account for generic Kubernetes resources.

configmaps

The instance manager can only read config maps that are related to the same cluster, such as custom monitoring queries

secrets

The instance manager can only read secrets that are related to the same cluster, namely: streaming replication user, application user, super user, LDAP authentication user, client CA, server CA, server certificate, backup credentials, custom monitoring queries

events

The instance manager can create an event for the cluster, informing the API server about a particular aspect of the PostgreSQL instance lifecycle

Here instead, we provide the same summary for resources specific to CloudNativePG.

clusters

The instance manager requires read-only permissions, namely get, list and watch, just for its own Cluster resource

clusters/status

The instance manager requires to update and patch the status of just its own Cluster resource

backups

The instance manager requires get and list permissions to read any Backup resource in the namespace. Additionally, it requires the delete permission to clean up the Kubernetes cluster by removing the Backup objects that do not have a counterpart in the object store - typically because of retention policies

backups/status

The instance manager requires to update and patch the status of any Backup resource in the namespace

Pod Security Policies

A Pod Security Policy is the Kubernetes way to define security rules and specifications that a pod needs to meet to run in a cluster. For InfoSec reasons, every Kubernetes platform should implement them.

CloudNativePG does not require privileged mode for containers execution. The PostgreSQL containers run as postgres system user. No component whatsoever requires running as root.

Likewise, Volumes access does not require privileges mode or root privileges either. Proper permissions must be properly assigned by the Kubernetes platform and/or administrators. The PostgreSQL containers run with a read-only root filesystem (i.e. no writable layer).

The operator explicitly sets the required security contexts.

Restricting Pod access using AppArmor

You can assign an AppArmor profile to the postgres, initdb, join, full-recovery and bootstrap-controller containers inside every Cluster pod through the container.apparmor.security.beta.kubernetes.io annotation.

    kind: Cluster
    metadata:
        name: cluster-apparmor
        annotations:
            container.apparmor.security.beta.kubernetes.io/postgres: runtime/default
            container.apparmor.security.beta.kubernetes.io/initdb: runtime/default
            container.apparmor.security.beta.kubernetes.io/join: runtime/default

The AppArmor configuration must be at Kubernetes node level, meaning that the underlying operating system must have this option enable and properly configured.

In case this is not the situation, and the annotations were added at the Cluster creation time, pods will not be created. On the other hand, if you add the annotations after the Cluster was created the pods in the cluster will be unable to start and you will get an error like this:

metadata.annotations[container.apparmor.security.beta.kubernetes.io/postgres]: Forbidden: may not add AppArmor annotations]

In such cases, please refer to your Kubernetes administrators and ask for the proper AppArmor profile to use.

Network Policies

The pods created by the Cluster resource can be controlled by Kubernetes network policies to enable/disable inbound and outbound network access at IP and TCP level. You can find more information in the networking document.

Network policies are beyond the scope of this document. Please refer to the "Network policies" section of the Kubernetes documentation for further information.

Exposed Ports

CloudNativePG exposes ports at operator, instance manager and operand levels, as listed in the table below:

System	Port number	Exposing	Name	TLS	Authentication
operator	9443	webhook server	webhook-server	Yes	Yes
operator	8080	metrics	metrics	No	No
instance manager	9187	metrics	metrics	Optional	No
instance manager	8000	status	status	Yes	No
operand	5432	PostgreSQL instance	postgresql	Optional	Yes

PostgreSQL

The current implementation of CloudNativePG automatically creates passwords and .pgpass files for the database owner and, only if requested by setting enableSuperuserAccess to true, for the postgres superuser.

As far as password encryption is concerned, CloudNativePG follows the default behavior of PostgreSQL: starting from PostgreSQL 14, password_encryption is by default set to scram-sha-256, while on earlier versions it is set to md5.

See the "Secrets" section in the "Connecting from an application" page for more information.

You can use those files to configure application access to the database.

By default, every replica is automatically configured to connect in physical async streaming replication with the current primary instance, with a special user called streaming_replica. The connection between nodes is encrypted and authentication is via TLS client certificates (please refer to the "Client TLS/SSL Connections" page for details). By default, the operator requires TLS v1.3 connections.

Currently, the operator allows administrators to add pg_hba.conf lines directly in the manifest as part of the pg_hba section of the postgresql configuration. The lines defined in the manifest are added to a default pg_hba.conf.

For further detail on how pg_hba.conf is managed by the operator, see the "PostgreSQL Configuration" page of the documentation.

The administrator can also customize the content of the pg_ident.conf file that by default only maps the local postgres user to the postgres user in the database.

For further detail on how pg_ident.conf is managed by the operator, see the "PostgreSQL Configuration" page of the documentation.

Storage

CloudNativePG delegates encryption at rest to the underlying storage class. For data protection in production environments, we highly recommend that you choose a storage class that supports encryption at rest.