Security

Breaking down a monolithic application into atomic services offers various benefits, including better agility, better scalability and better ability to reuse services. However, microservices also have particular security needs:

To defend against the man-in-the-middle attack, they need traffic encryption.
To provide flexible service access control, they need mutual TLS and fine-grained access policies.
To audit who did what at what time, they need auditing tools.

Istio Security tries to provide a comprehensive security solution to solve all these issues.

This page gives an overview on how you can use Istio security features to secure your services, wherever you run them. In particular, Istio security mitigates both insider and external threats against your data, endpoints, communication and platform.

The Istio security features provide strong identity, powerful policy, transparent TLS encryption, and authentication, authorization and audit (AAA) tools to protect your services and data. The goals of Istio security are:

Security by default: no changes needed for application code and infrastructure
Defense in depth: integrate with existing security systems to provide multiple layers of defense
Zero-trust network: build security solutions on untrusted networks

Visit our Mutual TLS Migration docs to start using Istio security features with your deployed services. Visit our Security Tasks for detailed instructions to use the security features.

High-level architecture

Security in Istio involves multiple components:

Citadel for key and certificate management
Sidecar and perimeter proxies to implement secure communication between clients and servers
Pilot to distribute authentication policies and secure naming information to the proxies
Mixer to manage authorization and auditing

In the following sections, we introduce the Istio security features in detail.

Istio identity

Identity is a fundamental concept of any security infrastructure. At the beginning of a service-to-service communication, the two parties must exchange credentials with their identity information for mutual authentication purposes. On the client side, the server's identity is checked against the secure naming information to see if it is an authorized runner of the service. On the server side, the server can determine what information the client can access based on the authorization policies, audit who accessed what at what time, charge clients based on the services they used, and reject any clients who failed to pay their bill from accessing the services.

In the Istio identity model, Istio uses the first-class service identity to determine the identity of a service. This gives great flexibility and granularity to represent a human user, an individual service, or a group of services. On platforms that do not have such identity available, Istio can use other identities that can group service instances, such as service names.

Istio service identities on different platforms:

Kubernetes: Kubernetes service account
GKE/GCE: may use GCP service account
GCP: GCP service account
AWS: AWS IAM user/role account
On-premises (non-Kubernetes): user account, custom service account, service name, istio service account, or GCP service account. The custom service account refers to the existing service account just like the identities that the customer's Identity Directory manages.

Istio security vs SPIFFE

The SPIFFE standard provides a specification for a framework capable of bootstrapping and issuing identities to services across heterogeneous environments.

Istio and SPIFFE share the same identity document: SVID (SPIFFE Verifiable Identity Document). For example, in Kubernetes, the X.509 certificate has the URI field in the format of “spiffe://<domain>/ns/<namespace>/sa/<serviceaccount>”. This enables Istio services to establish and accept connections with other SPIFFE-compliant systems.

Istio security and SPIRE, which is the implementation of SPIFFE, differ in the PKI implementation details. Istio provides a more comprehensive security solution, including authentication, authorization, and auditing.

PKI

The Istio PKI is built on top of Istio Citadel and securely provisions strong workload identities to every workload. Istio uses X.509 certificates to carry the identities in SPIFFE format. The PKI also automates the key & certificate rotation at scale.

Istio supports services running on both Kubernetes pods and on-premises machines. Currently we use different certificate key provisioning mechanisms for each scenario.

Kubernetes scenario

Citadel watches the Kubernetes apiserver, creates a SPIFFE certificate and key pair for each of the existing and new service accounts. Citadel stores the certificate and key pairs as Kubernetes secrets.
When you create a pod, Kubernetes mounts the certificate and key pair to the pod according to its service account via Kubernetes secret volume.
Citadel watches the lifetime of each certificate, and automatically rotates the certificates by rewriting the Kubernetes secrets.
Pilot generates the secure naming information, which defines what service account or accounts can run a certain service. Pilot then passes the secure naming information to the sidecar Envoy.

on-premises machines scenario

Citadel creates a gRPC service to take Certificate Signing Requests (CSRs).
Node agent generates a private key and CSR, and sends the CSR with its credentials to Citadel for signing.
Citadel validates the credentials carried with the CSR, and signs the CSR to generate the certificate.
The node agent sends both, the certificate received from Citadel and the private key, to Envoy.
The above CSR process repeats periodically for certificate and key rotation.

Node Agent in Kubernetes (in development)

In the near future, Istio will use node agent in Kubernetes for certificate and key provision, as shown in the figure below. Note that the identity provision flow for on-premises machines is the same so we only describe the Kubernetes scenario.

The flow goes as follows:

Citadel creates a gRPC service to take CSR requests.
Envoy sends a certificate and key request via Envoy secret discovery service (SDS) API.
Upon receiving the SDS request, node agent creates the private key and CSR, and sends the CSR with its credentials to Citadel for signing.
Citadel validates the credentials carried in the CSR, and signs the CSR to generate the certificate.
The node agent sends the certificate received from Citadel and the private key to Envoy, via the Envoy SDS API.
The above CSR process repeats periodically for certificate and key rotation.

Best practices

In this section, we provide a few deployment guidelines and discuss a real-world scenario.

Deployment guidelines

If there are multiple service operators (a.k.a. SREs) deploying different services in a medium- or large-size cluster, we recommend creating a separate Kubernetes namespace for each SRE team to isolate their access. For example, you can create a team1-ns namespace for team1, and team2-ns namespace for team2, such that both teams cannot access each other's services.

If Citadel is compromised, all its managed keys and certificates in the cluster may be exposed. We strongly recommend running Citadel in a dedicated namespace (for example, istio-citadel-ns), to restrict access to the cluster to only administrators.

Example

Let us consider a three-tier application with three services: photo-frontend, photo-backend, and datastore. The photo SRE team manages the photo-frontend and photo-backend services while the datastore SRE team manages the datastore service. The photo-frontend service can access photo-backend, and the photo-backend service can access datastore. However, the photo-frontend service cannot access datastore.

In this scenario, a cluster administrator creates three namespaces: istio-citadel-ns, photo-ns, and datastore-ns. The administrator has access to all namespaces and each team only has access to its own namespace. The photo SRE team creates two service accounts to run photo-frontend and photo-backend respectively in the photo-ns namespace. The datastore SRE team creates one service account to run the datastore service in the datastore-ns namespace. Moreover, we need to enforce the service access control in Istio Mixer such that photo-frontend cannot access datastore.

In this setup, Kubernetes can isolate the operator privileges on managing the services. Istio manages certificates and keys in all namespaces and enforces different access control rules to the services.

Authentication

Istio provides two types of authentication:

Transport authentication, also known as service-to-service authentication: verifies the direct client making the connection. Istio offers mutual TLS as a full stack solution for transport authentication. You can easily turn on this feature without requiring service code changes. This solution:
- Provides each service with a strong identity representing its role to enable interoperability across clusters and clouds.
- Secures service-to-service communication and end-user-to-service communication.
- Provides a key management system to automate key and certificate generation, distribution, and rotation.
Origin authentication, also known as end-user authentication: verifies the original client making the request as an end-user or device. Istio enables request-level authentication with JSON Web Token (JWT) validation and a streamlined developer experience for Auth0, Firebase Auth, Google Auth, and custom auth.

In both cases, Istio stores the authentication policies in the Istio config store via a custom Kubernetes API. Pilot keeps them up-to-date for each proxy, along with the keys where appropriate. Additionally, Istio supports authentication in permissive mode to help you understand how a policy change can affect your security posture before it becomes effective.

Mutual TLS authentication

Istio tunnels service-to-service communication through the client side and server side Envoy proxies. For a client to call a server, the steps followed are:

Istio re-routes the outbound traffic from a client to the client's local sidecar Envoy.
The client side Envoy starts a mutual TLS handshake with the server side Envoy. During the handshake, the client side Envoy also does a secure naming check to verify that the service account presented in the server certificate is authorized to run the target service.
The client side Envoy and the server side Envoy establish a mutual TLS connection, and Istio forwards the traffic from the client side Envoy to the server side Envoy.
After authorization, the server side Envoy forwards the traffic to the server service through local TCP connections.

Secure naming

The secure naming information contains N-to-N mappings from the server identities, which are encoded in certificates, to the service names that are referred by discovery service or DNS. A mapping from identity A to service name B means “A is allowed and authorized to run service B”. Pilot watches the Kubernetes apiserver, generates the secure naming information, and distributes it securely to the sidecar Envoys. The following example explains why secure naming is critical in authentication.

Suppose the legitimate servers that run the service datastore only use the infra-team identity. A malicious user has certificate and key for the test-team identity. The malicious user intends to impersonate the service to inspect the data sent from the clients. The malicious user deploys a forged server with the certificate and key for the test-team identity. Suppose the malicious user successfully hacked the discovery service or DNS to map the datastore service name to the forged server.

When a client calls the datastore service, it extracts the test-team identity from the server's certificate, and checks whether test-team is allowed to run datastore with the secure naming information. The client detects that test-team is not allowed to run the datastore service and the authentication fails.

Authentication architecture

You can specify authentication requirements for services receiving requests in an Istio mesh using authentication policies. The mesh operator uses .yaml files to specify the policies. The policies are saved in the Istio configuration storage once deployed. Pilot, the Istio controller, watches the configuration storage. Upon any policy changes, Pilot translates the new policy to the appropriate configuration telling the Envoy sidecar proxy how to perform the required authentication mechanisms. Pilot may fetch the public key and attach it to the configuration for JWT validation. Alternatively, Pilot provides the path to the keys and certificates the Istio system manages and installs them to the application pod for mutual TLS. You can find more info in the PKI section. Istio sends configurations to the targeted endpoints asynchronously. Once the proxy receives the configuration, the new authentication requirement takes effect immediately on that pod.

Client services, those that send requests, are responsible for following the necessary authentication mechanism. For origin authentication (JWT), the application is responsible for acquiring and attaching the JWT credential to the request. For mutual TLS, Istio provides a destination rule. The operator can use the destination rule to instruct client proxies to make initial connections using TLS with the certificates expected on the server side. You can find out more about how mutual TLS works in Istio in PKI and identity section.

Istio outputs identities with both types of authentication, as well as other claims in the credential if applicable, to the next layer: authorization. Additionally, operators can specify which identity, either from transport or origin authentication, should Istio use as ‘the principal’.

Authentication policies

This section provides more details about how Istio authentication policies work. As you'll remember from the Architecture section, authentication policies apply to requests that a service receives. To specify client-side authentication rules in mutual TLS, you need to specify the TLSSettings in the DestinationRule. You can find more information in our TLS settings reference docs. Like other Istio configuration, you can specify authentication policies in .yaml files. You deploy policies using kubectl.

The following example authentication policy specifies that transport authentication for the reviews service must use mutual TLS:

apiVersion: "authentication.istio.io/v1alpha1"
kind: "Policy"
metadata:
  name: "reviews"
spec:
  targets:
  - name: reviews
  peers:
  - mtls: {}

Policy storage scope

Istio can store authentication policies in namespace-scope or mesh-scope storage:

Mesh-scope policy is specified with a value of "MeshPolicy" for the kind field and the name "default". For example:

apiVersion: "authentication.istio.io/v1alpha1"
kind: "MeshPolicy"
metadata:
  name: "default"
spec:
  peers:
  - mtls: {}

Namespace-scope policy is specified with a value of "Policy" for the kind field and a specified namespace. If unspecified, the default namespace is used. For example for namespace ns1:
```
apiVersion: "authentication.istio.io/v1alpha1"
kind: "Policy"
metadata:
  name: "default"
  namespace: "ns1"
spec:
  peers:
  - mtls: {}
```

Policies in the namespace-scope storage can only affect services in the same namespace. Policies in mesh-scope can affect all services in the mesh. To prevent conflict and misuse, only one policy can be defined in mesh-scope storage. That policy must be named default and have an empty targets: section. You can find more information on our target selectors section.

Kubernetes currently implements the Istio configuration on Custom Resource Definitions (CRDs). These CRDs correspond to namespace-scope and cluster-scope CRDs and automatically inherit access protection via the Kubernetes RBAC. You can read more on the Kubernetes CRD documentation

Target selectors

An authentication policy's targets specify the service or services to which the policy applies. The following example shows a targets: section specifying that the policy applies to:

The product-page service on any port.
The reviews service on port 9000.

targets:
 - name: product-page
 - name: reviews
   ports:
   - number: 9000

If you don't provide a targets: section, Istio matches the policy to all services in the storage scope of the policy. Thus, the targets: section can help you specify the scope of the policies:

Mesh-wide policy: A policy defined in the mesh-scope storage with no target selector section. There can be at most one mesh-wide policy in the mesh.
Namespace-wide policy: A policy defined in the namespace-scope storage with name default and no target selector section. There can be at most one namespace-wide policy per namespace.
Service-specific policy: a policy defined in the namespace-scope storage, with non-empty target selector section. A namespace can have zero, one, or many service-specific policies.

For each service, Istio applies the narrowest matching policy. The order is: service-specific > namespace-wide > mesh-wide. If more than one service-specific policy matches a service, Istio selects one of them at random. Operators must avoid such conflicts when configuring their policies.

To enforce uniqueness for mesh-wide and namespace-wide policies, Istio accepts only one authentication policy per mesh and one authentication policy per namespace. Istio also requires mesh-wide and namespace-wide policies to have the specific name default.

Transport authentication

The peers: section defines the authentication methods and associated parameters supported for transport authentication in a policy. The section can list more than one method and only one method must be satisfied for the authentication to pass. However, as of the Istio 0.7 release, the only transport authentication method currently supported is mutual TLS. If you do not need transport authentication, skip this section entirely.

The following example shows the peers: section enabling transport authentication using mutual TLS.

peers:
  - mtls: {}

Currently, the mutual TLS setting doesn't require any parameters. Hence, -mtls: {}, - mtls: or - mtls: null declarations are treated the same. In the future, the mutual TLS setting may carry arguments to provide different mutual TLS implementations.

Origin authentication

The origins: section defines authentication methods and associated parameters supported for origin authentication. Istio only supports JWT origin authentication. However, a policy can list multiple JWTs by different issuers. Similar to peer authentication, only one of the listed methods must be satisfied for the authentication to pass.

The following example policy specifies an origins: section for origin authentication that accepts JWTs issued by Google:

origins:
- jwt:
    issuer: "https://accounts.google.com"
    jwksUri: "https://www.googleapis.com/oauth2/v3/certs"

Principal binding

The principal binding key-value pair defines the principal authentication for a policy. By default, Istio uses the authentication configured in the peers: section. If no authentication is configured in the peers: section, Istio leaves the authentication unset. Policy writers can overwrite this behavior with the USE_ORIGIN value. This value configures Istio to use the origin's authentication as the principal authentication instead. In future, we will support conditional binding, for example: USE_PEER when peer is X, otherwise USE_ORIGIN.

The following example shows the principalBinding key with a value of USE_ORIGIN:

principalBinding: USE_ORIGIN

Updating authentication policies

You can change an authentication policy at any time and Istio pushes the change to the endpoints almost in real time. However, Istio cannot guarantee that all endpoints receive a new policy at the same time. The following are recommendations to avoid disruption when updating your authentication policies:

To enable or disable mutual TLS: Use a temporary policy with a mode: key and a PERMISSIVE value. This configures receiving services to accept both types of traffic: plain text and TLS. Thus, no request is dropped. Once all clients switch to the expected protocol, with or without mutual TLS, you can replace the PERMISSIVE policy with the final policy. For more information, visit the Mutual TLS Migration tutorial.

peers:
- mTLS:
    mode: PERMISSIVE

For JWT authentication migration: requests should contain new JWT before changing policy. Once the server side has completely switched to the new policy, the old JWT, if there is any, can be removed. Client applications need to be changed for these changes to work.

Authorization

Istio's authorization feature - also known as Role-based Access Control (RBAC)

provides namespace-level, service-level, and method-level access control for services in an Istio Mesh. It features:
Role-Based semantics, which are simple and easy to use.
Service-to-service and end-user-to-service authorization.
Flexibility through custom properties support, for example conditions, in roles and role-bindings.
High performance, as Istio authorization is enforced natively on Envoy.

Authorization architecture

The above diagram shows the basic Istio authorization architecture. Operators specify Istio authorization policies using .yaml files. Once deployed, Istio saves the policies in the Istio Config Store.

Pilot watches for changes to Istio authorization policies. It fetches the updated authorization policies if it sees any changes. Pilot distributes Istio authorization policies to the Envoy proxies that are co-located with the service instances.

Each Envoy proxy runs an authorization engine that authorizes requests at runtime. When a request comes to the proxy, the authorization engine evaluates the request context against the current authorization policies, and returns the authorization result, ALLOW or DENY.

Enabling authorization

You enable Istio Authorization using a RbacConfig object. The RbacConfig object is a mesh-wide singleton with a fixed name value of default. You can only use one RbacConfig instance in the mesh. Like other Istio configuration objects, RbacConfig is defined as a Kubernetes CustomResourceDefinition (CRD) object.

In the RbacConfig object, the operator can specify a mode value, which can be:

OFF: Istio authorization is disabled.
ON: Istio authorization is enabled for all services in the mesh.
ON_WITH_INCLUSION: Istio authorization is enabled only for services and namespaces specified in the inclusion field.
ON_WITH_EXCLUSION: Istio authorization is enabled for all services in the mesh except the services and namespaces specified in the exclusion field.

In the following example, Istio authorization is enabled for the default namespace.

apiVersion: "rbac.istio.io/v1alpha1"
kind: RbacConfig
metadata:
  name: default
spec:
  mode: 'ON_WITH_INCLUSION'
  inclusion:
    namespaces: ["default"]

Authorization policy

To configure an Istio authorization policy, you specify a ServiceRole and ServiceRoleBinding. Like other Istio configuration objects, they are defined as Kubernetes CustomResourceDefinition (CRD) objects.

ServiceRole defines a group of permissions to access services.
ServiceRoleBinding grants a ServiceRole to particular subjects, such as a user, a group, or a service.

The combination of ServiceRole and ServiceRoleBinding specifies: who is allowed to do what under which conditions. Specifically:

who refers to the subjects section in ServiceRoleBinding.
what refers to the permissions section in ServiceRole.
which conditions refers to the conditions section you can specify with the Istio attributes in either ServiceRole or ServiceRoleBinding.

`ServiceRole`

A ServiceRole specification includes a list of rules, AKA permissions. Each rule has the following standard fields:

services: A list of service names. You can set the value to * to include all services in the specified namespace.
methods: A list of HTTP method names, for permissions on gRPC requests, the HTTP verb is always POST. You can set the value to * to include all HTTP methods.
paths: HTTP paths or gRPC methods. The gRPC methods must be in the form of /packageName.serviceName/methodName and are case sensitive.

A ServiceRole specification only applies to the namespace specified in the metadata section. The services and methods fields are required in a rule. paths is optional. If a rule is not specified or if it is set to *, it applies to any instance.

The example below shows a simple role: service-admin, which has full access to all services in the default namespace.

apiVersion: "rbac.istio.io/v1alpha1"
kind: ServiceRole
metadata:
  name: service-admin
  namespace: default
spec:
  rules:
  - services: ["*"]
    methods: ["*"]

Here is another role: products-viewer, which has read, "GET" and "HEAD", access to the service products.default.svc.cluster.local in the default namespace.

apiVersion: "rbac.istio.io/v1alpha1"
kind: ServiceRole
metadata:
  name: products-viewer
  namespace: default
spec:
  rules:
  - services: ["products.default.svc.cluster.local"]
    methods: ["GET", "HEAD"]

In addition, we support prefix matching and suffix matching for all the fields in a rule. For example, you can define a tester role with the following permissions in the default namespace:

Full access to all services with prefix "test-*", for example: test-bookstore, test-performance, test-api.default.svc.cluster.local.
Read ("GET") access to all paths with "*/reviews" suffix, for example: /books/reviews, /events/booksale/reviews, /reviews in service bookstore.default.svc.cluster.local.

apiVersion: "rbac.istio.io/v1alpha1"
kind: ServiceRole
metadata:
  name: tester
  namespace: default
spec:
  rules:
  - services: ["test-*"]
    methods: ["*"]
  - services: ["bookstore.default.svc.cluster.local"]
    paths: ["*/reviews"]
    methods: ["GET"]

In a ServiceRole, the combination of namespace + services + paths + methods defines how a service or services are accessed. In some situations, you may need to specify additional conditions for your rules. For example, a rule may only apply to a certain version of a service, or only apply to services with a specific label, like "foo". You can easily specify these conditions using constraints.

For example, the following ServiceRole definition adds a constraint that request.headers[version] is either "v1" or "v2" extending the previous products-viewer role. The supported key values of a constraint are listed in the constraints and properties page. In the case that the attribute is a map, for example request.headers, the key is an entry in the map, for example request.headers[version].

apiVersion: "rbac.istio.io/v1alpha1"
kind: ServiceRole
metadata:
  name: products-viewer-version
  namespace: default
spec:
  rules:
  - services: ["products.default.svc.cluster.local"]
    methods: ["GET", "HEAD"]
    constraints:
    - key: request.headers[version]
      values: ["v1", "v2"]

`ServiceRoleBinding`

A ServiceRoleBinding specification includes two parts:

roleRef refers to a ServiceRole resource in the same namespace.
A list of subjects that are assigned to the role.

You can either explicitly specify a subject with a user or with a set of properties. A property in a ServiceRoleBinding subject is similar to a constraint in a ServiceRole specification. A property also lets you use conditions to specify a set of accounts assigned to this role. It contains a key and its allowed values. The supported key values of a constraint are listed in the constraints and properties page.

The following example shows a ServiceRoleBinding named test-binding-products, which binds two subjects to the ServiceRole named "product-viewer" and has the following subjects

A service account representing service a, "service-account-a".
A service account representing the Ingress service "istio-ingress-service-account" and where the JWT email claim is "a@foo.com".

apiVersion: "rbac.istio.io/v1alpha1"
kind: ServiceRoleBinding
metadata:
  name: test-binding-products
  namespace: default
spec:
  subjects:
  - user: "service-account-a"
  - user: "istio-ingress-service-account"
    properties:
      request.auth.claims[email]: "a@foo.com"
  roleRef:
    kind: ServiceRole
    name: "products-viewer"

In case you want to make a service publicly accessible, you can set the subject to user: "*". This value assigns the ServiceRole to all (both authenticated and unauthenticated) users and services, for example:

apiVersion: "rbac.istio.io/v1alpha1"
kind: ServiceRoleBinding
metadata:
  name: binding-products-allusers
  namespace: default
spec:
  subjects:
  - user: "*"
  roleRef:
    kind: ServiceRole
    name: "products-viewer"

To assign the ServiceRole to only authenticated users and services, use source.principal: "*" instead, for example:

apiVersion: "rbac.istio.io/v1alpha1"
kind: ServiceRoleBinding
metadata:
  name: binding-products-all-authenticated-users
  namespace: default
spec:
  subjects:
  - properties:
      source.principal: "*"
  roleRef:
    kind: ServiceRole
    name: "products-viewer"

Using other authorization mechanisms

While we strongly recommend using the Istio authorization mechanisms, Istio is flexible enough to allow you to plug in your own authentication and authorization mechanisms via the Mixer component. To use and configure plugins in Mixer, visit our policies and telemetry adapters docs.