Rancher Government Solutions RKE2 Security Technical Implementation Guide

The Kubernetes Controller Manager is a background process that embeds core control loops regulating cluster system state through the API Server. Every process executed in a pod has an associated service account. By default, service accounts use the same credentials for authentication. Implementing the default settings poses a high risk to the Kubernetes Controller Manager. Setting the use-service-account-credential value lowers the attack surface by generating unique service accounts settings for each controller instance.

Once an attacker establishes access to a system, the attacker often attempts to create a persistent method of re-establishing access. One way to accomplish this is for the attacker to modify an existing account. Auditing of account creation is one method for mitigating this risk. A comprehensive account management process will ensure an audit trail documents the creation of application user accounts and, as required, notifies administrators and/or application when accounts are created. Such a process greatly reduces the risk that accounts will be surreptitiously created and provides logging that can be used for forensic purposes. Within Rancher RKE2, audit data can be generated from any of the deployed container platform components. This audit data is important when there are issues, such as security incidents, that must be investigated. To make the audit data worthwhile for the investigation of events, it is necessary to know where within the container platform the event occurred. To address access requirements, many application developers choose to integrate their applications with enterprise-level authentication/access/auditing mechanisms that meet or exceed access control policy requirements. Such integration allows the application developer to offload those access control functions and focus on core application features and functionality. Satisfies: SRG-APP-000026-CTR-000070, SRG-APP-000027-CTR-000075, SRG-APP-000028-CTR-000080, SRG-APP-000092-CTR-000165, SRG-APP-000095-CTR-000170, SRG-APP-000096-CTR-000175, SRG-APP-000097-CTR-000180, SRG-APP-000098-CTR-000185, SRG-APP-000099-CTR-000190, SRG-APP-000100-CTR-000195, SRG-APP-000101-CTR-000205, SRG-APP-000319-CTR-000745, SRG-APP-000320-CTR-000750, SRG-APP-000343-CTR-000780, SRG-APP-000358-CTR-000805, SRG-APP-000374-CTR-000865, SRG-APP-000375-CTR-000870, SRG-APP-000381-CTR-000905, SRG-APP-000409-CTR-000990, SRG-APP-000492-CTR-001220, SRG-APP-000493-CTR-001225, SRG-APP-000494-CTR-001230, SRG-APP-000495-CTR-001235, SRG-APP-000496-CTR-001240, SRG-APP-000497-CTR-001245, SRG-APP-000498-CTR-001250, SRG-APP-000499-CTR-001255, SRG-APP-000500-CTR-001260, SRG-APP-000501-CTR-001265, SRG-APP-000502-CTR-001270, SRG-APP-000503-CTR-001275, SRG-APP-000504-CTR-001280, SRG-APP-000505-CTR-001285, SRG-APP-000506-CTR-001290, SRG-APP-000507-CTR-001295, SRG-APP-000508-CTR-001300, SRG-APP-000509-CTR-001305, SRG-APP-000510-CTR-001310, SRG-APP-000516-CTR-000790, SRG-APP-00516-CTR-001325

Limiting the number of attack vectors and implementing authentication and encryption on the endpoints available to external sources is paramount when securing the overall Kubernetes cluster. The Controller Manager API service exposes port 10252/TCP by default for health and metrics information use. This port does not encrypt or authenticate connections. If this port is exposed externally, an attacker can use this port to attack the entire Kubernetes cluster. By setting the bind address to only localhost (i.e., 127.0.0.1), only those internal services that require health and metrics information can access the Control Manager API.

RKE2 registry is used to store images and is the keeper of truth for trusted images within the platform. To guarantee the images' integrity, access to the registry must be limited to those individuals who need to perform tasks to the images such as the update, creation, or deletion. Without this control access, images can be deleted that are in use by RKE2 causing a denial of service (DoS), and images can be modified or introduced without going through the testing and validation process allowing for the intentional or unintentional introduction of containers with flaws and vulnerabilities. By allowing anonymous connections, the controls put in place to secure the Kubelet can be bypassed. Setting anonymous authentication to "false" also disables unauthenticated requests from kubelets. While there are instances where anonymous connections may be needed (e.g., health checks) and Role-Based Access Controls (RBAC) are in place to limit the anonymous access, this access must be disabled and only enabled when necessary.

Ensure that the --audit-log-maxage argument is set to 30 or as appropriate. Retaining logs for at least 30 days ensures that you can go back in time and investigate or correlate any events. Set your audit log retention period to 30 days or as per your business requirements. Result: Pass

There are various configuration files, logs, access credentials, and other files stored on the host filesystem that contain sensitive information. These files could potentially put at risk, along with other specific workloads and components: - API server. - proxy. - scheduler. - controller. - etcd. - Kubernetes administrator account information. - audit log access, modification, and deletion. - application access, modification, and deletion. - container runtime files. If an attacker can gain access to these files, changes can be made to open vulnerabilities and bypass user authorizations inherent within Kubernetes with RBAC implemented. It is crucial to ensure user permissions are enforced down through to the operating system. Protecting file permissions will ensure that if a nonprivileged user gains access to the system they will still not be able to access protected information from the cluster API, cluster configuration, and sensitive cluster information. This control relies on the underlying operating system also having been properly configured to allow only least privileged access to perform required operations. Satisfies: SRG-APP-000133-CTR-000300, SRG-APP-000133-CTR-000295, SRG-APP-000133-CTR-000305, SRG-APP-000133-CTR-000310

It is important to disable any unnecessary components to reduce any potential attack surfaces. RKE2 allows disabling the following components: - rke2-canal - rke2-coredns - rke2-ingress-nginx - rke2-kube-proxy - rke2-metrics-server If utilizing any of these components presents a security risk, or if any of the components are not required then they can be disabled by using the "disable" flag. If any of the components are not required, they can be disabled by using the "disable" flag. Satisfies: SRG-APP-000141-CTR-000315, SRG-APP-000384-CTR-000915

Ports, protocols, and services within the RKE2 runtime must be controlled and conform to the PPSM CAL. Those ports, protocols, and services that fall outside the PPSM CAL must be blocked by the runtime. Instructions on the PPSM can be found in DOD Instruction 8551.01 Policy. RKE2 sets most ports and services configuration upon initiation; however, these ports can be changed after the fact to noncompliant configurations. It is important to verify core component configurations for compliance. API Server, Scheduler, Controller, ETCD, and User Pods should all be checked to ensure proper PPS configuration. Satisfies: SRG-APP-000142-CTR-000325, SRG-APP-000142-CTR-000330, SRG-APP-000383-CTR-000910

Secrets, such as passwords, keys, tokens, and certificates should not be stored as environment variables. These environment variables are accessible inside RKE2 by the "Get Pod" API call, and by any system, such as CI/CD pipeline, which has access to the definition file of the container. Secrets must be mounted from files or stored within password vaults.

Terminating an idle session within a short time period reduces the window of opportunity for unauthorized personnel to take control of a management session enabled on the console or console port that has been left unattended. In addition, quickly terminating an idle session will also free up resources committed by the managed network element. Terminating network connections associated with communications sessions includes, for example, de-allocating associated TCP/IP address/port pairs at the operating system level, or de-allocating networking assignments at the application level if multiple application sessions are using a single, operating-system-level network connection. This does not mean that the application terminates all sessions or network access; it only ends the inactive session and releases the resources associated with that session.

RKE2 runs as isolated as possible. RKE2 is a container-based Kubernetes distribution. A container image is essentially a complete and executable version of an application, which relies only on the host's OS kernel. Running containers use resource isolation features in the OS kernel, such as cgroups in Linux, to run multiple independent containers on the same OS. Unless part of the core RKE2 system or configured explicitly, containers managed by RKE2 should not have access to host resources. Proper hardening of the surrounding environment is independent of RKE2 but ensures overall security stature. When Kubernetes launches a container, there are several mechanisms available to ensure complete deployments: - When a primary container process fails it is destroyed rebooted. - When Liveness checks fail for the container deployment it is destroyed rebooted. - If a readiness check fails at any point after the deployment the container is destroyed rebooted. - Kubernetes has the ability to rollback a deployment configuration to a previous state if a deployment fails. - Failover traffic to a working replica if any of the previous problems are encountered. System kernel is responsible for memory, disk, and task management. The kernel provides a gateway between the system hardware and software. Kubernetes requires kernel access to allocate resources to the Control Plane. Threat actors that penetrate the system kernel can inject malicious code or hijack the Kubernetes architecture. It is vital to implement protections through Kubernetes components to reduce the attack surface.

Separating user functionality from management functionality is a requirement for all the components within the Kubernetes Control Plane. Without the separation, users may have access to management functions that can degrade the Kubernetes architecture and the services being offered, and can offer a method to bypass testing and validation of functions before introduced into a production environment. Satisfies: SRG-APP-000243-CTR-000600, SRG-APP-000431-CTR-001065, SRG-APP-000211-CTR-000530, SRG-APP-000243-CTR-000595

Admission controllers intercept requests to the Kubernetes API before an object is instantiated. Enabling the admissions webhook allows for Kubernetes to apply policies against objects that are to be created, read, updated or deleted. Admissions controllers can be used for: - Prevent pod’s ability to run privileged containers - Prevent pod’s ability to use privileged escalation - Controlling pod’s access to volume types - Controlling pod’s access to host file system - Controlling pod’s usage of host networking objects and configuration Satisfies: SRG-APP-000340-CTR-000770, SRG-APP-000342-CTR-000775

Controlling access to those users and roles responsible for patching and updating RKE2 reduces the risk of untested or potentially malicious software from being installed within the platform. This access may be separate from the access required to install container images into the registry and those access requirements required to instantiate an image into a service. Explicit privileges (escalated or administrative privileges) provide the regular user with explicit capabilities and control that exceeds the rights of a regular user. Kubernetes uses the API Server to control communication to the other services that makeup Kubernetes. The use of authorizations and not the default of "AlwaysAllow" enables the Kubernetes functions control to only the groups that need them. To control access, the API server must have one of the following options set for the authorization mode: --authorization-mode=ABAC Attribute-Based Access Control (ABAC) mode allows a user to configure policies using local files. --authorization-mode=RBAC Role-based access control (RBAC) mode allows a user to create and store policies using the Kubernetes API. --authorization-mode=Webhook WebHook is an HTTP callback mode that allows a user to manage authorization using a remote REST endpoint. --authorization-mode=Node Node authorization is a special-purpose authorization mode that specifically authorizes API requests made by kubelets. --authorization-mode=AlwaysDeny This flag blocks all requests. Use this flag only for testing. Satisfies: SRG-APP-000378-CTR-000880, SRG-APP-000378-CTR-000885

Encrypting secrets at rest in etcd. By default, RKE2 will create an encryption key and configuration file and pass these to the Kubernetes API server. The result is that RKE2 automatically encrypts Kubernetes Secret objects when writing them to etcd.

Previous versions of Rancher RKE2 components that are not removed after updates have been installed may be exploited by adversaries by causing older components to execute which contain vulnerabilities. When these components are deleted, the likelihood of this happening is removed.

Software supporting RKE2, images in the registry must stay up to date with the latest patches, service packs, and hot fixes. Not updating RKE2 and container images will expose the organization to vulnerabilities. Flaws discovered during security assessments, continuous monitoring, incident response activities, or information system error handling must also be addressed expeditiously. Organization-defined time periods for updating security-relevant container platform components may vary based on a variety of factors including, for example, the security category of the information system or the criticality of the update (i.e., severity of the vulnerability related to the discovered flaw). This requirement will apply to software patch management solutions used to install patches across the enclave and to applications themselves that are not part of that patch management solution. For example, many browsers today provide the capability to install their own patch software. Patch criticality, as well as system criticality will vary. Therefore, the tactical situations regarding the patch management process will also vary. This means that the time period utilized must be a configurable parameter. Time frames for application of security-relevant software updates may be dependent upon the Information Assurance Vulnerability Management (IAVM) process. RKE2 components will be configured to check for and install security-relevant software updates within an identified time period from the availability of the update. RKE2 registry will ensure the images are current. The specific time period will be defined by an authoritative source (e.g., IAVM, CTOs, DTMs, and STIGs).