普通程序员看k8s基于角色的访问控制（RBAC）

一、知识准备

● 上一节描述了k8s的账户管理，本文描述基于角色的访问控制

● 网上RBAC的文章非常多，具体概念大神们也解释得很详细，本文没有站在高屋建瓴的角度去描述RBAC，而是站在一个普通程序员的视角，去看待RBAC

● 我理解的基于角色的访问控制，如图：

                                        +------------+
                          +------------>|all:resource|-+
                          |             +------------+ |       +----------------+
                     +----+---+                        |       |                |
+------+             |role:   |                        +------>+  namespaces    |
|user1 |------------>|admin   |                                |                |
+------+             +--------+         +------------+         |  pod           |
                                  +---->|get:pod     |-------->+                |
                                  |     +------------+         |  service       |
                                  |                            |                |
+------+             +--------+   |                            |  deployment    |
|user2 |------------>|role:   |---+                            |                |
+------+             |guest   |         ....                   |  daemonsets    |
                     +--------+                                |                |
                                                               |  ...           |
                                        +------------+         |                |
                                        |list:service|         +----------------+
                                        +------------+

账户：请求kube-api必要的身份验证。身份验证之后只是被允许进入集群，但是不一定有访问资源的权限

角色：账户登录之后扮演的角色

规则：在k8s集群内允许的操作。比如get namespace、get pod等

资源：k8s各种各样的资源

角色绑定：资源与规则组成访问权限，将角色绑定在访问权限上，最后账户登录之后扮演不同的角色

二、环境准备

组件	版本
OS	Ubuntu 18.04.1 LTS
docker	18.06.0-ce
k8s	v1.10.1

三、创建角色

角色分为role与clusterrole，区别在于前者是namespace级别，后者是整个k8s cluster级别

（1）创建role

创建一个default namespace的普通角色，权限是对pod的get、list、watch

apiVersion: rbac.authorization.k8s.io/v1
kind: Role
metadata:
  name: po_r
  namespace: default
rules:
- apiGroups:
  - ""
  resources:
  - pods
  verbs:
  - get
  - list
  - watch

（2）创建clusterrole

创建一个具有pod的get、list、watch的clusterrole，这个角色是cluster级别的，也就是所有的namespace

apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  name: cluster_po_r
rules:
- apiGroups:
  - ""
  resources:
  - pods
  verbs:
  - get
  - list
  - watch

四、userAccount角色绑定

（1）role绑定

先将上一节的mrvolleyball账号绑定一个namespace级别的role，使用rolebinding

apiVersion: rbac.authorization.k8s.io/v1
kind: RoleBinding
metadata:
  name: mrvolleyball
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: Role
  name: po_r
subjects:
- apiGroup: rbac.authorization.k8s.io
  kind: User
  name: mrvolleyball

再看mrvolleyball的账户，已经拥有了对default namespace的get pod权限

root@k8s-master:~# kubectl get pod
NAME                                      READY     STATUS    RESTARTS   AGE
busybox                                   1/1       Running   89         3d
nginx-deployment-77d7f55978-lvfkf         1/1       Running   0          21h

不过他既没有service的访问权限，更没有别的namespace的访问权限

root@k8s-master:~# kubectl get svc
Error from server (Forbidden): services is forbidden: User "mrvolleyball" cannot list services in the namespace "default"
root@k8s-master:~# kubectl get pod -n kube-system
Error from server (Forbidden): pods is forbidden: User "mrvolleyball" cannot list pods in the namespace "kube-system"

（2）clusterrole绑定

绑定clusterrole，让mrvollerball能够读取所有namespace的pod

apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: mrvolleyball
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: ClusterRole
  name: cluster_po_r
subjects:
- apiGroup: rbac.authorization.k8s.io
  kind: User
  name: mrvolleyball

这时候mrvolleyball也可以读取其他namespace的pod

root@k8s-master:~# kubectl get pod -n kube-system
NAME                                        READY     STATUS             RESTARTS   AGE
calico-kube-controllers-55c94797d7-vr9k6    1/1       Running            0          15d
calico-node-86xgg                           1/1       Running            0          7d
calico-node-drsl2                           1/1       Running            0          7d
calico-node-l67h4                           1/1       Running            0          7d
coredns-794b5f8589-vgclz                    1/1       Running            4809       15d
coredns-794b5f8589-xzpxd                    1/1       Running            0          130d
default-http-backend-6cb5c56987-tcx92       1/1       Running            0          139d
kubernetes-dashboard-77c46fc984-djdd4       1/1       Running            0          139d
nginx-ingress-controller-6d95468cc8-ft6nn   1/1       Running            0          140d

这里需要注意几种资源的绑定：

rolebinding --> role

clusterrolebinding --> clusterrole

rolebinding --> clusterrole

● 第一种是namespace级别的绑定，只对role所在的namespace生效

● 第二种是cluster级别的绑定，对于集群所有的namespace都生效

● 第三种是namespace级别的绑定，只对role所在的namespace生效

有同学要问，第一种和第三种有什么区别：

apiVersion: rbac.authorization.k8s.io/v1
kind: RoleBinding
metadata:
  name: mrvolleyball
  namespace: default
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: ClusterRole
  name: cluster_po_r
subjects:
- apiGroup: rbac.authorization.k8s.io
  kind: User
  name: mrvolleyball

区别主要是为了方便管理：

● 比如有10个namespace，分配给了10个不同的团队，那每个团队都需要使用自己的userAccount操作自己所在的namespace

● 最简单的做法，就是在10个不同namespace创建10个role，然后通过rolebinding将role绑定userAccount。如果需要添加userAccount，那这个管理成本就会升高

● 这时候创建一个clusterrole，由于他们对各自的namespace的行为都是一样的，那就只需要通过rolebinding将clusterrole绑定在userAccount，只需要创建1次clusterrole

五、serviceAccount角色绑定

serviceAccount要提供给pod内访问，或者是外部服务的访问。为了验证serviceAccount，我们来模拟pod去访问api

首先先创建一个serviceAccount

root@k8s-master:/etc/kubernetes/ssl# kubectl create serviceaccount helloworld
serviceaccount "helloworld" created
root@k8s-master:/etc/kubernetes/ssl# kubectl get sa helloworld -o yaml
apiVersion: v1
kind: ServiceAccount
metadata:
  creationTimestamp: 2019-03-12T06:36:00Z
  name: helloworld
  namespace: default
  resourceVersion: "24381027"
  selfLink: /api/v1/namespaces/default/serviceaccounts/helloworld
  uid: 1987b4e6-4491-11e9-96bc-525400c34cf1
secrets:
- name: helloworld-token-sr9xb

serviceAccount获取默认生成一个token，该token是用于pod访问kube-api的身份验证（详情访问上一篇文章）

root@k8s-master:/etc/kubernetes/ssl# kubectl get secret helloworld-token-sr9xb -o yaml
apiVersion: v1
data:
...
  token: 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
kind: Secret
metadata:
  annotations:
    kubernetes.io/service-account.name: helloworld
    kubernetes.io/service-account.uid: 1987b4e6-4491-11e9-96bc-525400c34cf1
  creationTimestamp: 2019-03-12T06:36:00Z
  name: helloworld-token-sr9xb
  namespace: default
  resourceVersion: "24381026"
  selfLink: /api/v1/namespaces/default/secrets/helloworld-token-sr9xb
  uid: 198c824f-4491-11e9-96bc-525400c34cf1
type: kubernetes.io/service-account-token

拿到该token，将其转码之后作为连接kube-api的身份验证

root@k8s-master:/etc/kubernetes/ssl# token=`kubectl get secret helloworld-token-sr9xb -o yaml | grep "token:" | awk '{print $2}' | base64 -d`
root@k8s-master:/etc/kubernetes/ssl# kubectl config set-credentials helloworld --token=$token
User "helloworld" set.

配置rolebinding，po_r这个role只能对default namespace的pod有读权限

apiVersion: rbac.authorization.k8s.io/v1
kind: RoleBinding
metadata:
  name: helloworld-role-binding
  namespace: default
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: Role
  name: po_r
subjects:
- kind: ServiceAccount
  name: helloworld
  namespace: default

尝试访问一下：

root@k8s-master:~# kubectl config use-context context@helloworld
Switched to context "context@helloworld".
root@k8s-master:~# kubectl get pod
NAME                                      READY     STATUS    RESTARTS   AGE
busybox                                   1/1       Running   187        7d

该serviceAccount也只有default namespace的访问权限

root@k8s-master:~# kubectl get pod -n kube-system
Error from server (Forbidden): pods is forbidden: User "system:serviceaccount:default:helloworld" cannot list pods in the namespace "kube-system"

六、小结

● 本文介绍了k8s基于权限的访问，详细k8s当中role、clusterrole与rolebinding、clusterrolebinding

● 并且模拟了不同类型的账户访问kube-api时不同的认证方式

---

至此，本文结束

在下才疏学浅，有撒汤漏水的，请各位不吝赐教...