Movatterモバイル変換


[0]ホーム

URL:


Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

Sign up
Appearance settings

Provides different Google Service Accounts and Scopes for pods running on Kubernetes

License

NotificationsYou must be signed in to change notification settings

imduffy15/k8s-gke-service-account-assigner

Repository files navigation

Build StatusGitHub tagDocker PullsGo Report Card

k8s-gke-service-account-assigner

Provides Google Service Account Tokens to containers running inside a kubernetes cluster based on annotations.

Inspired by and heavily based offkube2iam

Context

Service accounts are attached to instances and are accessible by services through the transparentusage by the google-cloud-sdk of the Google instance metadata API. When using the google-cloud-sdk,a call is made to the Google instance metadata API which provides temporary credentialsthat are then used to make calls to the Google service.

Problem statement

The problem is that in a multi-tenanted containers based world, multiple containers will be sharing the underlyingnodes. Given containers will share the same underlying nodes, they each get the same Google service account credentials.

Solution

The solution is to redirect the traffic that is going to the Google instance metadata API for docker containers to a containerrunning on each instance, make a call to the Google IAM Credentials API to retrieve temporary credentials and return these to the caller.Other calls will be proxied to the Google instance metadata API. This container will need to run with host networking enabledso that it can call the Google instance metadata API itself.

Usage

Service accounts

It is necessary to create an service account which has the roleroles/iam.serviceAccountTokenCreator so it get tokens for other service accounts and can assign it to each pod.

This service account should be associated to the kubernetes cluster and the kubernetes cluster should have the scopehttps://www.googleapis.com/auth/cloud-platform so that it can query theIAM credentials API

Additionally, its necessary to have enabled theGoogle IAM Credentials API for your project.

k8s-gke-service-account-assigner daemonset

Run the k8s-gke-service-account-assigner container as a daemonset (so that it runs on each worker) withhostNetwork: true.The k8s-gke-service-account-assigner daemon and iptables rule (see below) need to run before all other pods that would requireaccess to Google resources.

apiVersion:extensions/v1beta1kind:DaemonSetmetadata:name:saassignerlabels:app:saassignerspec:template:metadata:labels:name:saassignerspec:hostNetwork:truecontainers:        -image:imduffy15/k8s-gke-service-account-assigner:latestname:saassignerargs:            -"--iptables=true"            -"--host-ip=$(HOST_IP)"            -"--node=$(NODE_NAME)"            -"--default-service-account=default"            -"--default-scopes=https://www.googleapis.com/auth/devstorage.read_write,https://www.googleapis.com/auth/logging.write,https://www.googleapis.com/auth/monitoring,https://www.googleapis.com/auth/servicecontrol,https://www.googleapis.com/auth/service.management.readonly,https://www.googleapis.com/auth/trace.append"env:            -name:HOST_IPvalueFrom:fieldRef:fieldPath:status.podIP            -name:NODE_NAMEvalueFrom:fieldRef:fieldPath:spec.nodeNameports:            -containerPort:8181hostPort:8181name:http

iptables

To prevent containers from directly accessing the Google instance metadata API and gaining unwanted access to Google resources,the traffic to169.254.169.254 must be proxied for docker containers.

iptables \  --append PREROUTING \  --protocol tcp \  --destination 169.254.169.254 \  --dport 80 \  --in-interface eth0 \  --jump DNAT \  --table nat \  --to-destination`curl http://169.254.169.254/computeMetadata/v1/instance/network-interfaces/0/ip -H"Metadata-Flavor: Google"`:8181

This rule can be added automatically by setting--iptables=true, setting theHOST_IP environmentvariable, and running the container in a privileged security context (or more granularly withCAP_NET_ADMIN andCAP_NET_RAW).

kubernetes annotation

Add anaccounts.google.com/service-account andaccounts.google.com/scopes annotation to your pods with the service accountand scopes that you want to used for this pod.

apiVersion:v1kind:Podmetadata:name:debug-shelllabels:name:debug-shellannotations:accounts.google.com/service-account:"<PROJECT-ID>-compute@developer.gserviceaccount.com"accounts.google.com/scopes:"https://www.googleapis.com/auth/cloud-platform"spec:restartPolicy:Nevercontainers:    -image:imduffy15/docker-gcloudimagePullPolicy:Alwaysname:debug-shelltty:true

You can use--default-service-account and--default-scopes to set a fallback service account and scope to use when annotation is not set.

Namespace Restrictions

By using the flag --namespace-restrictions you can enable a mode in which the roles that pods can assume is restrictedby an annotation on the pod's namespace. This annotation should be in the form of a json array.

To allow the debug-shell pod specified above to run in the default namespace your namespace would look like the following.

apiVersion:v1kind:Namespacemetadata:annotations:accounts.google.com/allowed-service-accounts:|      ["<PROJECT-ID>-compute@developer.gserviceaccount.com"]name:default

The entries in the array may also beglob patterns.For example, you could match any user-managed service account for a project using["*@<PROJECT-ID>.iam.gserviceaccount.com"].

RBAC Setup

This is the basic RBAC setup to get k8s-gke-service-account-assigner working correctly when your cluster is using rbac. Below is the bare minimum to get k8s-gke-service-account-assigner working.

First we need to make a service account.

---apiVersion:v1kind:ServiceAccountmetadata:name:k8s-gke-service-account-assignernamespace:kube-system

Next we need to setup roles and binding for the the process.

---apiVersion:v1items:  -apiVersion:rbac.authorization.k8s.io/v1beta1kind:ClusterRolemetadata:name:k8s-gke-service-account-assignerrules:      -apiGroups:[""]resources:["namespaces","pods"]verbs:["get","watch","list"]  -apiVersion:rbac.authorization.k8s.io/v1beta1kind:ClusterRoleBindingmetadata:name:k8s-gke-service-account-assignersubjects:    -kind:ServiceAccountname:k8s-gke-service-account-assignernamespace:kube-systemroleRef:kind:ClusterRolename:k8s-gke-service-account-assignerapiGroup:rbac.authorization.k8s.iokind:List

You will notice this lives in the kube-system namespace to allow for easier seperation between system services and other services.

Debug

By using the --debug flag you can enable some extra features making debugging easier:

  • /debug/store endpoint enabled to dump knowledge of namespaces and service account association.

Options

By default,k8s-gke-service-account-assigner will use the in-cluster method to connect to the kubernetes master, and use theaccounts.google.com/service-account andaccounts.google.com/scopes annotations to retrieve the serviceaccount and scopes for the container.

$ k8s-gke-service-account-assigner --helpUsage of ./k8s-gke-service-account-assigner:      --api-server string                   Endpointfor the api server      --api-token string                    Token to authenticate with the api server      --app-port string                     Http port (default"8181")      --backoff-max-elapsed-time duration   Max elapsedtimefor backoff when queryingfor role. (default 2s)      --backoff-max-interval duration       Max intervalfor backoff when queryingfor role. (default 1s)      --debug                               Enable debug features      --default-scopes string               Fallback scopes to use when annotation is notset      --default-service-account string      Fallback service account to use when annotation is notset      --enable-metadata-proxy               Send traffic to next-hop proxy      --host-interface string               Host interfacefor proxying google compute engine metadata (default"eth0")      --host-ip string                      IP address of host      --iam-role-key string                 Pod annotation key used to retrieve the IAM role (default"accounts.google.com/service-account")      --insecure                            Kubernetes server should be accessed without verifying the TLS. Testing only      --iptables                            Add iptables rule (also requires --host-ip)      --log-format string                   Log format (text/json) (default"text")      --log-level string                    Log level (default"info")      --metadata-addr string                Addressfor the google compute engine metadata (default"169.254.169.254")      --metadata-proxy-addr string          Addressfor the next-hop proxy, defaults to GKE's metadata-proxy location (default "127.0.0.1:988")      --namespace-key string                Namespace annotation key used to retrieve the service accounts allowed (value in annotation should be json array) (default "accounts.google.com/allowed-service-accounts")      --namespace-restrictions              Enable namespace restrictions      --node string                         Name of the node where k8s-gke-service-account-assigner is running      --verbose                             Verbose      --version                             Print the version and exits

Usage with GKE Metadata Concealment

GKE's Metadata concealment protects some potentially sensitive system metadata from user workloads running on your cluster. It does so by deployingk8s-metadata-proxy and routing traffic to metadata service via this proxy. K8s-gke-service-account-assigner will take precedence, by putting its iptables rule before the k8s-metadata-proxy one, and therefore effectively bypassing it. If you want to use both, set--enable-metadata-proxy, in which case the traffic will we sent to the--metadata-proxy-addr (defaults to127.0.0.1:988, GKE's metadata-proxy default location).

Development loop

  • Create a Google Kubernetes Engine cluster
  • Run skaffold dev
  • Run make watch
  • Create a container and exec onto it to make queries against the deployed k8s-gke-service-account-assigner instance

About

Provides different Google Service Accounts and Scopes for pods running on Kubernetes

Topics

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Contributors5


[8]ページ先頭

©2009-2025 Movatter.jp