kube-rbac-proxy

NOTE: This project is alpha stage. Flags, configuration, behavior and design may change significantly in following releases.

The kube-rbac-proxy is a small HTTP proxy for a single upstream, that can perform RBAC authorization against the Kubernetes API using SubjectAccessReviews.

In Kubernetes clusters without NetworkPolicies any Pod can perform requests to every other Pod in the cluster. This proxy was developed in order to restrict requests to only those Pods, that present a valid and RBAC authorized token or client TLS certificate.

Usage

The kube-rbac-proxy has all glog flags for logging purposes. To use the kube-rbac-proxy there are a few flags you may want to set:

• --upstream: This is the upstream you want to proxy to.
• --resource-attributes-file: This file specifies details on the SubjectAccessReview you want to be performed on a request. For example, this could contain that an entity performing a request has to be allowed to perform a get on the Deployment called my-frontend-app.

See the examples/ directory for the following examples:

• non-resource-url example
• resource-attributes example

All command line flags:

$ kube-rbac-proxy -h
Usage of _output/linux/amd64/kube-rbac-proxy:
      --alsologtostderr                             log to standard error as well as files
      --auth-header-fields-enabled                  When set to true, kube-rbac-proxy adds auth-related fields to the headers of http requests sent to the upstream
      --auth-header-groups-field-name string        The name of the field inside a http(2) request header to tell the upstream server about the user's groups (default "x-remote-groups")
      --auth-header-groups-field-separator string   The separator string used for concatenating multiple group names in a groups header field's value (default "|")
      --auth-header-user-field-name string          The name of the field inside a http(2) request header to tell the upstream server about the user's name (default "x-remote-user")
      --client-ca-file string                       If set, any request presenting a client certificate signed by one of the authorities in the client-ca-file is authenticated with an identity corresponding to the CommonName of the client certificate.
      --insecure-listen-address string              The address the kube-rbac-proxy HTTP server should listen on.
      --log_backtrace_at traceLocation              when logging hits line file:N, emit a stack trace (default :0)
      --log_dir string                              If non-empty, write log files in this directory
      --logtostderr                                 log to standard error instead of files
      --resource-attributes-file string             File spec of attributes-record to use for SubjectAccessReview. If unspecified, requests will attempted to be verified through non-resource-url attributes in the SubjectAccessReview.
      --secure-listen-address string                The address the kube-rbac-proxy HTTPs server should listen on.
      --stderrthreshold severity                    logs at or above this threshold go to stderr (default 2)
      --tls-cert-file string                        File containing the default x509 Certificate for HTTPS. (CA cert, if any, concatenated after server cert)
      --tls-cipher-suites stringSlice               Comma-separated list of cipher suites for the server. Values are from tls package constants (https://golang.org/pkg/crypto/tls/#pkg-constants). If omitted, the default Go cipher suites will be used
      --tls-min-version string                      Minimum TLS version supported. Value must match version names from https://golang.org/pkg/crypto/tls/#pkg-constants. (default "VersionTLS12")
      --tls-private-key-file string                 File containing the default x509 private key matching --tls-cert-file.
      --upstream string                             The upstream URL to proxy to once requests have successfully been authenticated and authorized.
      --upstream-force-h2c                          Force h2c to communiate with the upstream. This is required when the upstream speaks h2c(http/2 cleartext - insecure variant of http/2) only. For example, go-grpc server in the insecure mode, such as helm's tiller w/o TLS, speaks h2c only
  -v, --v Level                                     log level for V logs
      --vmodule moduleSpec                          comma-separated list of pattern=N settings for file-filtered logging

Why?

You may ask yourself, why not just use the Kubernetes apiserver proxy functionality? There are two reasons why this makes sense, the first is to take load off of the Kubernetes API, so it can be used for actual requests serving the cluster components, rather than in order to serve client requests. The second and more important reason is, this proxy is intended to be a sidecar that accepts incoming HTTP requests. This way, one can ensure that a request is truly authorized, instead of being able to access an application simply because an entity has network access to it.

Motivation

I developed this proxy in order to be able to protect Prometheus metrics endpoints. In a scenario, where an attacker might obtain full control over a Pod, that attacker would have the ability to discover a lot of information about the workload as well as the current load of the respective workload. This information could originate for example from the node-exporter and kube-stat-metrics. Both of those metric sources can commonly be found in Prometheus monitoring stacks on Kubernetes.

This project was created to specifically solve the above problem, however, I felt there is a larger need for such a proxy in general.

How does it work?

On an incoming request, kube-rbac-proxy first figures out which user is performing the request. The kube-rbac-proxy supports using client TLS certificates, as well as tokens. In case of a client certificates, the certificate is simply validated against the configured CA. In case of a bearer token being presented, the authentication.k8s.io is used to perform a TokenReview.

Once a user has been authenticated, again the authentication.k8s.io is used to perform a SubjectAccessReview, in order to authorize the respective request, to ensure the authenticated user has the required RBAC roles.

Notes on ServiceAccount token security

Note that when using tokens for authentication, the receiving side can use the token to impersonate the client. Only use token authentication, when the receiving side is already higher privileged or the token itself is super low privileged, such as when the only roles binded to it are for authorization purposes with this project. Passing around highly privileged tokens is a security risk, and is not recommended.

This project was built to be used to protect metrics of cluster components. These cluster components are much higher privileged than the Prometheus Pod, so if those Pods were to use the token provided by Prometheus it would actually be lower privileged. It is not recommended to use this method for non infrastructure components.

For better security properties use mTLS for authentication instead, and for user authentication, other methods have yet to be added.

Why are NetworkPolicies not enough?

There are a couple of reasons why the existance of NetworkPolicies may not cover the same use case(s):

• NetworkPolicies are not available in all providers, installers and distros.
• NetworkPolicies do not apply to Pods with HostNetworking enabled, the use case I created this project with the Prometheus node-exporter requires this.
• Once TLS/OIDC is supported, the kube-rbac-proxy can be used to perform AuthN/AuthZ on users.

Differentiation to Envoy/Istio

This projects is not intended to compete with Envoy or IstioMesh. Although on the surface they seem similar, the goals and usage complement each other. It's perfectly ok to use Envoy as the ingress point of traffic of a Pod, which then forwards traffic to the kube-rbac-proxy, which in turn then proxies to the actually serving application.

Additionally, to my knowledge Envoy neither has nor plans Kubernetes specific RBAC/AuthZ support (maybe it shouldn’t even). My knowledge may very well be incomplete, please point out if it is. After all I'm happy if I don't have to maintain more code, but as long as this serves a purpose to me and no other project can provide it, I'll maintain this.

Roadmap

PR are more than welcome!

• Tests
• OIDC support