Skip to content

Enforcing egress traffic using Istio’s authorization policies📜

An Istio Egress gateway is just another envoy instance similar to the Ingress but with the purpose to control outbound traffic. Istio uses ingress and egress gateways to configure load balancers executing at the edge of a service mesh. An ingress gateway allows you to define entry points into the mesh that all incoming traffic flows through. Egress gateway is a symmetrical concept; it defines exit points from the mesh. Egress gateways allow you to apply Istio features, for example, monitoring and route rules, to traffic exiting the mesh.

This article describes how to enforce outbound authorization policies using Istio’s Egress gateway in a similar matter when enforcing inbound policies. For this we use the sleep service in two separate namespaces within the mesh to access external services at Google and Yahoo.

Note

One important consideration to be aware of is that Istio cannot securely enforce that all egress traffic actually flows through the egress gateways. Istio only enables such flow through its sidecar proxies. If attackers bypass the sidecar proxy, they could directly access external services without traversing the egress gateway. Kubernetes network policies can be used to prevent outbound traffic at the cluster level, for more information read here and here.


Prep-work📜

  1. Install istio: See our guide to standup a minimal cluster with Istio on your VM BigBang Quickstart

  2. Add an egress gateway instance by editing your values file. Add the following:

    istio:
      postRenderers:
        - kustomize:
            patchesStrategicMerge:
            # Change operator deployment to be a rolling update
            - kind: IstioOperator
              apiVersion: install.istio.io/v1alpha1
              metadata:
                name: istiocontrolplane
                namespace: istio-system
              spec:
                components:
                  egressGateways:
                  - name: istio-egressgateway
                    enabled: true
    

Tip

By default, Istio configures the handling of proxy accessing external services by using the outboundTrafficPolicy option. ALLOW_ANY is the default value to enabling access to outbound services and REGISTRY_ONLY gets the sidecar proxies to restrict access if the host is not defined in the service registry using the ServiceEntry resource.

Deploy the sleep service📜

Label the namespace for sidecar injection:

kubectl label ns default istio-injection=enabled

  1. Deploy sleep to the default namespace.

    sleep.yaml
     1
     2
     3
     4
     5
     6
     7
     8
     9
    10
    11
    12
    13
    14
    15
    16
    17
    18
    19
    20
    21
    22
    23
    24
    25
    26
    27
    28
    29
    30
    31
    32
    33
    34
    35
    36
    37
    38
    39
    40
    41
    42
    43
    44
    45
    46
    47
    48
    49
    50
    51
    52
    53
    54
    55
    56
    57
    58
    59
    60
    61
    62
    63
    64
    65
    66
    # Copyright Istio Authors
    #
    #   Licensed under the Apache License, Version 2.0 (the "License");
    #   you may not use this file except in compliance with the License.
    #   You may obtain a copy of the License at
    #
    #       http://www.apache.org/licenses/LICENSE-2.0
    #
    #   Unless required by applicable law or agreed to in writing, software
    #   distributed under the License is distributed on an "AS IS" BASIS,
    #   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    #   See the License for the specific language governing permissions and
    #   limitations under the License.
    
    ##################################################################################################
    # Sleep service
    ##################################################################################################
    apiVersion: v1
    kind: ServiceAccount
    metadata:
      name: sleep
    ---
    apiVersion: v1
    kind: Service
    metadata:
      name: sleep
      labels:
        app: sleep
        service: sleep
    spec:
      ports:
      - port: 80
        name: http
      selector:
        app: sleep
    ---
    apiVersion: apps/v1
    kind: Deployment
    metadata:
      name: sleep
    spec:
      replicas: 1
      selector:
        matchLabels:
          app: sleep
      template:
        metadata:
          labels:
            app: sleep
        spec:
          terminationGracePeriodSeconds: 0
          serviceAccountName: sleep
          containers:
          - name: sleep
            image: curlimages/curl
            command: ["/bin/sleep", "infinity"]
            imagePullPolicy: IfNotPresent
            volumeMounts:
            - mountPath: /etc/sleep/tls
              name: secret-volume
          volumes:
          - name: secret-volume
            secret:
              secretName: sleep-secret
              optional: true
    ---
    
    kubectl apply -f sleep.yaml
    
  2. Deploy the sleep service in the otherns namespace

kubectl create ns otherns

Label the namespace for sidecar injection:

kubectl label ns otherns istio-injection=enabled

kubectl apply -n otherns -f sleep.yaml

Tip

If you run into image pull errors, you’ll have to create a private-registry secret if deploying via BB, when pulling images from registry1.

Export sleep pods name into variables📜

export SLEEP_POD1=$(kubectl get pod -l app=sleep -ojsonpath='{.items[0].metadata.name}')
export SLEEP_POD2=$(kubectl get pod -n otherns -l app=sleep -ojsonpath='{.items[0].metadata.name}')

Test sleep accessing Google and Yahoo📜

kubectl exec $SLEEP_POD1 -it -- curl -I https://developers.google.com

You should expect a similar response like:

HTTP/2 200
last-modified: Mon, 18 Apr 2022 19:50:38 GMT
content-type: text/html; charset=utf-8
set-cookie: _ga_devsite=GA1.3.17352200.1651777078; Expires=Sat, 04-May-2024 18:57:58 GMT; Max-Age=63072000; Path=/
content-security-policy: base-uri 'self'; object-src 'none'; script-src 'strict-dynamic' 'unsafe-inline' https: http: 'nonce-6YT4DgbNb9SFKpYNAAh6BVQ1HrIWUp' 'unsafe-eval'; report-uri https://csp.withgoogle.com/csp/devsite/v2
strict-transport-security: max-age=63072000; includeSubdomains; preload
x-frame-options: SAMEORIGIN
x-xss-protection: 0
x-content-type-options: nosniff
cache-control: no-cache, must-revalidate
expires: 0
pragma: no-cache
x-cloud-trace-context: 3943a8b1bdf28d721eae4f82696ba2c4
content-length: 142275
date: Thu, 05 May 2022 18:57:58 GMT
server: Google Frontend

Now the other service:

kubectl exec $SLEEP_POD2 -n otherns -it -- curl -I https://developer.yahoo.com

You should expect a similar response like:

HTTP/2 200
referrer-policy: no-referrer-when-downgrade
strict-transport-security: max-age=15552000
x-frame-options: SAMEORIGIN
x-powered-by: Express
cache-control: private, max-age=0, no-cache
content-security-policy-report-only: default-src 'none'; connect-src 'self' *.yimg.com https://www.google-analytics.com *.yahoo.com *.doubleclick.net; font-src 'self' *.bootstrapcdn.com; frame-src 'self' *.soundcloud.com *.twitter.com; img-src 'self' data: *.yimg.com https://www.google-analytics.com *.yahoo.com https://www.google.com/ads/ga-audiences *.pendo.io *.twitter.com *.twimg.com; script-src 'self' 'nonce-25FqRrNIte3nmHy7Es/O4Q==' *.yimg.com https://www.google-analytics.com https://ssl.google-analytics.com *.github.com/flurrydev/ *.pendo.io *.twitter.com *.twimg.com; style-src 'self' 'unsafe-inline' *.yimg.com *.twitter.com *.twimg.com https://github.githubassets.com/assets/ *.bootstrapcdn.com; report-uri /csp-report
content-type: text/html; charset=utf-8
content-length: 61158
etag: W/"eee6-355CS9JqgK79WnB2sdI2zK9AvBw"
vary: Accept-Encoding
date: Thu, 05 May 2022 19:00:06 GMT
x-envoy-upstream-service-time: 2315
server: ATS
age: 3
expect-ct: max-age=31536000, report-uri="http://csp.yahoo.com/beacon/csp?src=yahoocom-expect-ct-report-only"
x-xss-protection: 1; mode=block
x-content-type-options: nosniff

If you want you can test the other other address on the other sleep pod. We can confirm the pods have outbound access to Google and Yahoo.

Restrict outbound access📜

We need to modify the outbound traffic policy from ALLOW_ANY to REGISTRY_ONLY which enforces that only hosts defined with ServiceEntry resources are part of the mesh service registry; could be accessed by sidecars of the mesh:

  1. Change the outboundTrafficPolicy by adjusting your values file:
    istio:
      values:
        meshConfig:
          outboundTrafficPolicy:
            mode: REGISTRY_ONLY
    

Test sleep access again📜

kubectl exec $SLEEP_POD1 -it -- curl -I https://developers.google.com

You should expect a similar response like:

curl: (35) OpenSSL SSL_connect: SSL_ERROR_SYSCALL in connection to developers.google.com:443
command terminated with exit code 35

Now the other service:

kubectl exec $SLEEP_POD2 -n otherns -it -- curl -I https://developer.yahoo.com

You should expect a similar response like:

curl: (35) OpenSSL SSL_connect: SSL_ERROR_SYSCALL in connection to developer.yahoo.com:443
command terminated with exit code 35

The error is due to the enforcement of the new policy that only allows outbound traffic to services that are part of the registry.

Note

There could be a slight delay on the configuration being propagated to the sidecars. During this time, they still allow access to the external services.


Add the Google and Yahoo services to the mesh service registry📜

Our Google ServiceEntry looks like this:

apiVersion: networking.istio.io/v1beta1
kind: ServiceEntry
metadata:
  name: external-developers-google-com
spec:
  hosts:
  - developers.google.com
  exportTo:
  - "."
  location: MESH_EXTERNAL
  resolution: DNS
  ports:
  - number: 443
    name: https
    protocol: HTTPS
  - number: 80
    name: http
    protocol: HTTP

Apply the resource:

kubectl apply -f serviceentry-google.yaml

Note

Notice the exportTo: - "." section of the service entry resource specifying that is only applicable to the current namespace where applied. You can also change this to "*" for all namespaces in the mesh.

Test access to the service:

kubectl exec $SLEEP_POD1 -it -- curl -I https://developers.google.com

You should expect a 200 response code now. But what if we test this sleep service to Yahoo?

kubectl exec $SLEEP_POD1 -it -- curl -I https://developer.yahoo.com

You should expect an error along the lines:

curl: (35) OpenSSL SSL_connect: Connection reset by peer in connection to developer.yahoo.com:443
command terminated with exit code 35

This is because we only allowed outbound traffic to Google from the default namespace where the SLEEP_POD1 lives. Any outbound traffic from SLEEP_POD2 should still be restricted, lets enabled traffic to Google:

kubectl apply -n otherns -f serviceentry-google.yaml

You should expect a 200 response code from both pods:

kubectl exec $SLEEP_POD2 -n otherns -it -- curl -I https://developers.google.com
kubectl exec $SLEEP_POD1 -it -- curl -I https://developers.google.com

Notice how Yahoo is still restricted on both services:

kubectl exec $SLEEP_POD1 -it -- curl -I https://developer.yahoo.com
kubectl exec $SLEEP_POD2 -n otherns -it -- curl -I https://developer.yahoo.com

Take a look at the Yahoo ServiceEntry:

apiVersion: networking.istio.io/v1beta1
kind: ServiceEntry
metadata:
  name: external-developer-yahoo-com
spec:
  hosts:
  - developer.yahoo.com
  exportTo:
  - "."
  location: MESH_EXTERNAL
  resolution: DNS
  ports:
  - number: 443
    name: https
    protocol: HTTPS
  - number: 80
    name: http
    protocol: HTTP

Enable traffic on the default namespace and test it:

kubectl apply -f serviceentry-yahoo.yaml
kubectl exec $SLEEP_POD1 -it -- curl -I https://developer.yahoo.com

Now Yahoo on the otherns namespace:

kubectl apply -n otherns -f serviceentry-yahoo.yaml
kubectl exec $SLEEP_POD2 -n otherns -it -- curl -I https://developer.yahoo.com

You should expect a 200 response code from both pods. Any other request to other external hosts that are not Yahoo or Google should be restricted and only allowed if in the service registry from the default and otherns namespaces.

Notice how when the ServiceEntry resource is created in the target namespace, external communication to the defined host is allowed only from the sidecar proxies on that namespace.

Cleanup📜

kubectl delete -f serviceentry-google.yaml
kubectl delete -n otherns -f serviceentry-google.yaml
kubectl delete -f serviceentry-yahoo.yaml
kubectl delete -n otherns -f serviceentry-yahoo.yaml

Enforcing egress traffic using authorization policies📜

So far by changing the outbound traffic policy to REGISTRY_ONLY we can enforce how our proxy sidecars allow outbound traffic from the mesh to the external hosts only defined with our ServiceEntry resources, but we don’t have a fine-grained control with them.

Using the service entries is more like a opening/closing a “faucet” in the namespace and having to create resources per namespace will create a maintenance burden. You can change the resource to be scoped for all namespaces (“*”) and not just the target namespace but just with the ServiceEntry resource you can’t control which workload within the namespace can or cannot access an external host.

We can accomplish this fine-grained control with an AuthorizationPolicy after we flow internally originated outbound traffic to the Egress gateway making act as a proxy with the help of VirtualService, Gateway, DestinationRule resources along with ServiceEntrys on how outbound traffic should flow.

In a similar manner when dealing with inbound traffic routing, we can create a DestinationRule that flows traffic from the sidecars to the egress gateway and a second DestinationRule that flows the traffic to the external host.

These DestinationRules are bound to a VirtualService that matches traffic to the whole mesh Gateway and the Gateway defined for the external host. By doing this setup, we can rely on the ServiceEntry and AuthorizationPolicy resources to ensure that only allowed/denied outbound traffic defined for namespaces or principals (k8s ServiceAccount) can reach the external hosts.

Note

Notice this example relies on Istio’s automatic mutual TLS. This means services within the mesh send encrypted TLS traffic, but traffic external to the mesh may or may not be encrypted. This is called SIMPLE TLS mode. To force traffic to be TLS encrypted when leaving the mesh, known as TLS origination, we use MUTUAL TLS mode. This can be setup in the DestinationRule by defining the secret name that holds the client credentials certificate. See more details here.


Route internal outbound traffic to the egress gateway📜

After deleting the ServiceEntry resources used on the previous section, make sure your mesh is still restricting outbound access. Verify that there are no other resources that can conflict with the configuration like DestinationRule, VirtualService, Gateway or AuthorizationPolicy resources:

kubectl exec $SLEEP_POD1 -it -- curl -I https://developers.google.com
kubectl exec $SLEEP_POD2 -n otherns -it -- curl -I https://developers.google.com
kubectl exec $SLEEP_POD1 -it -- curl -I https://developer.yahoo.com
kubectl exec $SLEEP_POD2 -n otherns -it -- curl -I https://developer.yahoo.com

For all requests expect an error along the lines:

curl: (35) OpenSSL SSL_connect: Connection reset by peer in connection to developer.yahoo.com:443
command terminated with exit code 35

Analyze the following resources external-google.yaml and external-yahoo.yaml:

Google:

apiVersion: networking.istio.io/v1alpha3
kind: ServiceEntry
metadata:
  name: google
spec:
  hosts:
  - developers.google.com
  ports:
  - number: 80
    name: http
    protocol: HTTP
  - number: 443
    name: https
    protocol: HTTPS
  resolution: DNS
---
apiVersion: networking.istio.io/v1alpha3
kind: Gateway
metadata:
  name: istio-google-egressgateway
spec:
  selector:
    istio: egressgateway # use Istio default gateway implementation
  servers:
  - port:
      number: 80
      name: https-port-for-tls-origination
      protocol: HTTPS
    hosts:
    - developers.google.com
    tls:
      mode: ISTIO_MUTUAL
---
# Routes internal outbound traffic to the egress gateway using Istio's mTLS
apiVersion: networking.istio.io/v1alpha3
kind: DestinationRule
metadata:
  name: egressgateway-for-google
spec:
  host: istio-egressgateway.istio-system.svc.cluster.local
  subsets:
  - name: google
    trafficPolicy:
      portLevelSettings:
      - port:
          number: 80
        tls:
          mode: ISTIO_MUTUAL
          sni: developers.google.com
---
apiVersion: networking.istio.io/v1alpha3
kind: VirtualService
metadata:
  name: direct-google-through-egress-gateway
spec:
  hosts:
  - developers.google.com
  gateways:
  - istio-google-egressgateway
  - mesh
  # route HTTP traffic to developers.google.com through the egress gateway for the entire mesh
  http:
  - match:
    - gateways:
      - mesh # apply to sidecars in the mesh
      port: 80
    route:
    - destination:
        host: istio-egressgateway.istio-system.svc.cluster.local
        subset: google
        port:
          number: 80
  # at the egress gateway, route developers.google.com to the real destination outside the mesh
  - match:
    - gateways:
      - istio-google-egressgateway
      port: 80
    route:
    - destination:
        host: developers.google.com
        port:
          number: 443
      weight: 100
---
apiVersion: networking.istio.io/v1alpha3
kind: DestinationRule
metadata:
  name: originate-tls-for-developers-google-com
spec:
  host: developers.google.com
  trafficPolicy:
    loadBalancer:
      simple: ROUND_ROBIN
    portLevelSettings:
    - port:
        number: 443
      tls:
        mode: SIMPLE # initiates HTTPS for connections to developers.google.com

Yahoo:

apiVersion: networking.istio.io/v1alpha3
kind: ServiceEntry
metadata:
  name: yahoo
spec:
  hosts:
  - developer.yahoo.com
  ports:
  - number: 80
    name: http
    protocol: HTTP
  - number: 443
    name: https
    protocol: HTTPS
  resolution: DNS
---
apiVersion: networking.istio.io/v1alpha3
kind: Gateway
metadata:
  name: istio-yahoo-egressgateway
spec:
  selector:
    istio: egressgateway # use Istio default gateway implementation
  servers:
  - port:
      number: 80
      name: https-port-for-tls-origination
      protocol: HTTPS
    hosts:
    - developer.yahoo.com
    tls:
      mode: ISTIO_MUTUAL
---
# Routes internal outbound traffic to the egress gateway using Istio's mTLS
apiVersion: networking.istio.io/v1alpha3
kind: DestinationRule
metadata:
  name: egressgateway-for-yahoo
spec:
  host: istio-egressgateway.istio-system.svc.cluster.local
  subsets:
  - name: yahoo
    trafficPolicy:
      portLevelSettings:
      - port:
          number: 80
        tls:
          mode: ISTIO_MUTUAL
          sni: developer.yahoo.com
---
apiVersion: networking.istio.io/v1alpha3
kind: VirtualService
metadata:
  name: direct-yahoo-through-egress-gateway
spec:
  hosts:
  - developer.yahoo.com
  gateways:
  - istio-yahoo-egressgateway
  - mesh
  # route HTTP traffic to developers.google.com through the egress gateway for the entire mesh
  http:
  - match:
    - gateways:
      - mesh # apply to sidecars in the mesh
      port: 80
    route:
    - destination:
        host: istio-egressgateway.istio-system.svc.cluster.local
        subset: yahoo
        port:
          number: 80
  # at the egress gateway, route developer.yahoo.com to the real destination outside the mesh
  - match:
    - gateways:
      - istio-yahoo-egressgateway
      port: 80
    route:
    - destination:
        host: developer.yahoo.com
        port:
          number: 443
      weight: 100
---
apiVersion: networking.istio.io/v1alpha3
kind: DestinationRule
metadata:
  name: originate-tls-for-developer-yahoo-com
spec:
  host: developer.yahoo.com
  trafficPolicy:
    loadBalancer:
      simple: ROUND_ROBIN
    portLevelSettings:
    - port:
        number: 443
      tls:
        mode: SIMPLE # initiates HTTPS for connections to developer.yahoo.com

In the previous resources you can find:

  • ServiceEntry resources to enable external access to Google and Yahoo
  • Gateway resources for each host configuring the egress gateway instance for originating traffic to the external host
  • VirtualService resources for each host bound for the entire mesh. This includes the configuration for matching traffic from within the mesh (sidecars) and routing it through the egress gateway and outbound to the external host.
  • DestinationRule resources applied to the traffic after being routed by the VirtualService. The first rule defines internal traffic using the sni (hostname) and sets up Istio’s automatic mTLS: ISTIO_MUTUAL. The second DestinationRule defines how to initiate HTTPS connections to the external host.

Apply these resources and test accessing the services:

kubectl apply -f external-google.yaml -n istio-system
kubectl apply -f external-yahoo.yaml -n istio-system

Note

This time we are applying all these resources on the istio-system namespace where the egress gateway instance resides. This is with the intention to easily manage egress traffic where the egress gateway instance resides, facilitating the management of the AuthorizationPolicys.


Tip

If you run into issues like 503 errors, make sure you disable BigBang networkPolicies by doing:

networkPolicies:
  enabled: false

Access developers.google.com:

kubectl exec $SLEEP_POD1 -it -- curl -I http://developers.google.com

Expect a 200 response along the lines:

HTTP/1.1 200 OK
last-modified: Mon, 18 Apr 2022 19:50:38 GMT
content-type: text/html; charset=utf-8
set-cookie: _ga_devsite=GA1.3.878699971.1652214977; Expires=Thu, 09-May-2024 20:36:17 GMT; Max-Age=63072000; Path=/
content-security-policy: base-uri 'self'; object-src 'none'; script-src 'strict-dynamic' 'unsafe-inline' https: http: 'nonce-yp4hjMbNOIwavPWy28V4k9lOdtSb6X' 'unsafe-eval'; report-uri https://csp.withgoogle.com/csp/devsite/v2
strict-transport-security: max-age=63072000; includeSubdomains; preload
x-frame-options: SAMEORIGIN
x-xss-protection: 0
x-content-type-options: nosniff
cache-control: no-cache, must-revalidate
expires: 0
pragma: no-cache
x-cloud-trace-context: 61c3fe6ffc0bc6bc209d455b04d9d86e
content-length: 142287
date: Tue, 10 May 2022 20:36:17 GMT
server: envoy
x-envoy-upstream-service-time: 420

Tip

If you run into issues like 503 errors, make sure you disable BigBang networkPolicies by doing:

networkPolicies:
  enabled: false

Tail the logs of the istio-proxy sidecar:

kubectl logs $SLEEP_POD1 -f -c istio-proxy

Expect an entry from the sidecar to the egress:

[2022-05-10T20:36:16.973Z] "HEAD / HTTP/1.1" 200 - via_upstream - "-" 0 0 421 420 "-" "curl/7.83.0-DEV" "5ab0ed38-2e77-92a8-bb44-0a07573cd530" "developers.google.com" "10.100.2.6:8080" outbound|80|google|istio-egressgateway.istio-system.svc.cluster.local 10.100.0.5:48236 173.194.217.101:80 10.100.0.5:48764 - -

Tail the logs of the egressgateway:

export EG_POD=$(kubectl get pod -l istio=egressgateway -ojsonpath='{.items[0].metadata.name}' -n istio-system)
kubectl logs $EG_POD -f -n istio-system

Expect an entry from the egress to the external host:

[2022-05-10T20:36:16.981Z] "HEAD / HTTP/2" 200 - via_upstream - "-" 0 0 395 394 "10.100.0.5" "curl/7.83.0-DEV" "5ab0ed38-2e77-92a8-bb44-0a07573cd530" "developers.google.com" "173.194.217.101:443" outbound|443||developers.google.com 10.100.2.6:51492 10.100.2.6:8080 10.100.0.5:48236 developers.google.com -

Tip

Notice how the internal outbound traffic is intentionally originated using http in order to rely on Istio’s automatic mTLS within the mesh. Then using the DestinationRule TLS mode SIMPLE, the egress instance does a secure request to the external host.


Repeat the same steps using the sleep service on the otherns for the Yahoo host:

kubectl exec $SLEEP_POD2 -n otherns -it -- curl -I http://developer.yahoo.com

Expect an entry like the following on the sidecar logs:

[2022-05-10T20:51:37.091Z] "HEAD / HTTP/1.1" 200 - via_upstream - "-" 0 0 2389 2389 "-" "curl/7.83.0-DEV" "b2e32c17-2db4-925f-bbd7-c201b549f7ef" "developer.yahoo.com" "10.100.2.6:8080" outbound|80|yahoo|istio-egressgateway.istio-system.svc.cluster.local 10.100.1.6:38682 69.147.92.11:80 10.100.1.6:47940 - -

And on the egress:

[2022-05-10T20:51:37.099Z] "HEAD / HTTP/2" 200 - via_upstream - "-" 0 0 2364 2363 "10.100.1.6" "curl/7.83.0-DEV" "b2e32c17-2db4-925f-bbd7-c201b549f7ef" "developer.yahoo.com" "69.147.92.11:443" outbound|443||developer.yahoo.com 10.100.2.6:40486 10.100.2.6:8080 10.100.1.6:38682 developer.yahoo.com -

At this time you can test the other external host on the opposite sleep service and notice is still accessible:

kubectl exec $SLEEP_POD2 -n otherns -it -- curl -I http://developers.google.com
kubectl exec $SLEEP_POD1 -it -- curl -I http://developer.yahoo.com

Expect 200 responses from either sleep service.

Enforce authorization polices📜

Although we can enforce denying external access by removing ServiceEntry resources while the REGISTRY_ONLY mode is active, we can also do it with a more fine-grained control using AuthorizationPolicys after the correct configuration is in place.

Take a look at this policy that allows no traffic out:

apiVersion: security.istio.io/v1beta1
kind: AuthorizationPolicy
metadata:
 name: allow-nothing
 namespace: istio-system
spec:
  {}

Apply the authz-policy-allow-nothing.yaml file that enforces this purpose:

kubectl apply -f authz-policy-allow-nothing.yaml

Try to access the services again:

kubectl exec $SLEEP_POD1 -it -- curl -I http://developer.yahoo.com
kubectl exec $SLEEP_POD2 -n otherns -it -- curl -I http://developers.google.com
kubectl exec $SLEEP_POD1 -it -- curl -I http://developers.google.com
kubectl exec $SLEEP_POD2 -n otherns -it -- curl -I http://developer.yahoo.com

Expect a response along the lines:

HTTP/1.1 403 Forbidden
content-length: 19
content-type: text/plain
date: Tue, 10 May 2022 21:08:13 GMT
server: envoy
x-envoy-upstream-service-time: 13

Note

Keep in mind some requests could be allowed while the configuration takes place. If your request is allowed, wait a minute and try again.

Now apply the authz-policy-allow-google.yaml allowing the default ns to do requests to developers.google.com:

authz-policy-allow-google.yaml
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
apiVersion: security.istio.io/v1beta1
kind: AuthorizationPolicy
metadata:
  name: external-allow-developers-google-com
  # No ns means that applies to all ns in a mesh
spec:
  # allow-list for the identities that can call the host
  action: ALLOW
  rules:
  - from:
    - source:
        namespaces: ["default"]
    to:
    - operation:
        methods: ["GET"]
        hosts:
        - developers.google.com
kubectl apply -f authz-policy-allow-google.yaml

Now test again:

kubectl exec $SLEEP_POD1 -it -- curl -I http://developers.google.com

Notice that even when applying the authz-policy-allow-google.yaml allowing the default ns to do requests to developers.google.com it still gets forbidden. This is because in AuthorizationPolicys the DENY action is evaluated before the ALLOW one.

Delete the resources:

kubectl delete authorizationpolicies.security.istio.io -n istio-system allow-nothing
kubectl delete authorizationpolicies.security.istio.io -n default external-allow-developers-google-com

Enforce policies per namespace📜

For this use case we allow the sleep service on the default namespace to access google but not yahoo and the for the sleep service on the otherns namespace it allows yahoo but not google.

Analyze the following policies:

apiVersion: security.istio.io/v1beta1
kind: AuthorizationPolicy
metadata:
  name: external-deny-developers-google-com
  # No ns means that applies to all ns in a mesh
spec:
  # allow-list for the identities that can call the host
  action: DENY
  rules:
  - from:
    - source:
        namespaces: ["otherns"]
    to:
    - operation:
        hosts:
        - developers.google.com
apiVersion: security.istio.io/v1beta1
kind: AuthorizationPolicy
metadata:
  name: external-deny-developer-yahoo-com
  # No ns means that applies to all ns in a mesh
spec:
  # allow-list for the identities that can call the host
  action: DENY
  rules:
  - from:
    - source:
        namespaces: ["default"]
    to:
    - operation:
        hosts:
        - developer.yahoo.com

Apply the following policies:

kubectl apply -f authz-policy-deny-google.yaml -n istio-system
kubectl apply -f authz-policy-deny-yahoo.yaml -n istio-system

Try to access the services again:

kubectl exec $SLEEP_POD1 -it -- curl -I http://developer.yahoo.com
kubectl exec $SLEEP_POD2 -n otherns -it -- curl -I http://developers.google.com
kubectl exec $SLEEP_POD1 -it -- curl -I http://developers.google.com
kubectl exec $SLEEP_POD2 -n otherns -it -- curl -I http://developer.yahoo.com

For the first couple requests expect a 403 Forbidden response and for the last couple expect a 200 response.

Tail the logs for the egress gateway and expect an entry describing the policy matched:

[2022-05-10T21:51:49.396Z] "HEAD / HTTP/2" 403 - rbac_access_denied_matched_policy[ns[istio-system]-policy[external-deny-developer-yahoo-com]-rule[0]] - "-" 0 0 0 - "10.100.0.6" "curl/7.83.0-DEV" "a503ba03-f09a-914d-85a4-995a0c1d5b16" "developer.yahoo.com" "-" outbound|443||developer.yahoo.com - 10.100.2.6:8080 10.100.0.6:58370 developer.yahoo.com -

Delete the resources:

kubectl delete authorizationpolicies.security.istio.io -n istio-system external-deny-developer-yahoo-com external-deny-developers-google-com

Enforce policies per workload using service account principals📜

Using account principals provides the fine-grained control we need per workload within a namespace. For this use case deploy another set of sleep services on the otherns namespace:

kubectl apply -f sleep-custom.yaml -n otherns

The yaml file above is the traditional sleep service adding custom names for identification.

This will create the sleep-google and sleep-yahoo services beside the existing one. Save the pods names:

export SLEEP_POD_G=$(kubectl get pod -n otherns -l app=sleep-google -ojsonpath='{.items[0].metadata.name}')
export SLEEP_POD_Y=$(kubectl get pod -n otherns -l app=sleep-yahoo -ojsonpath='{.items[0].metadata.name}')

Apply the following policies that block the sleep-google service to access Yahoo and sleep-yahoo service to access Google within the otherns namespace, still leaving access to both hosts from the sleep service:

apiVersion: security.istio.io/v1beta1
kind: AuthorizationPolicy
metadata:
  name: external-deny-developers-google-com
  # No ns means that applies to all ns in a mesh
spec:
  # allow-list for the identities that can call the host
  action: DENY
  rules:
  - from:
    - source:
        principals: ["cluster.local/ns/otherns/sa/sleep-yahoo"]
    when:
    - key: connection.sni
      values:
      - developers.google.com
apiVersion: security.istio.io/v1beta1
kind: AuthorizationPolicy
metadata:
  name: external-deny-developer-yahoo-com
  # No ns means that applies to all ns in a mesh
spec:
  # allow-list for the identities that can call the host
  action: DENY
  rules:
  - from:
    - source:
        principals: ["cluster.local/ns/otherns/sa/sleep-google"]
    when:
    - key: connection.sni
      values:
      - developer.yahoo.com

kubectl apply -f authz-policy-deny-google-custom.yaml -n istio-system
kubectl apply -f authz-policy-deny-yahoo-custom.yaml -n istio-system

Test the policies:

kubectl exec $SLEEP_POD_G -n otherns -it -- curl -I http://developers.google.com
kubectl exec $SLEEP_POD_G -n otherns -it -- curl -I http://developer.yahoo.com
kubectl exec $SLEEP_POD_Y -n otherns -it -- curl -I http://developers.google.com
kubectl exec $SLEEP_POD_Y -n otherns -it -- curl -I http://developer.yahoo.com
kubectl exec $SLEEP_POD2 -n otherns -it -- curl -I http://developers.google.com
kubectl exec $SLEEP_POD2 -n otherns -it -- curl -I http://developer.yahoo.com

Note

The second and third responses should be 403 forbidden as they are denied by the policies. The rest should return a 200.

You successfully used AuthorizationPolicys to enforce outbound traffic through the egress gateway.

Clean up:

kubectl delete authorizationpolicies -n istio-system external-deny-developers-google-com external-deny-developer-yahoo-com
kubectl delete se -n istio-system google yahoo
kubectl delete vs -n istio-system direct-google-through-egress-gateway direct-yahoo-through-egress-gateway
kubectl delete dr -n istio-system egressgateway-for-google originate-tls-for-developers-google-com egressgateway-for-yahoo originate-tls-for-developer-yahoo-com
kubectl delete gateway -n istio-system istio-google-egressgateway istio-yahoo-egressgateway


Enforce policies per workload using service account principals and TLS originated traffic📜

This is really similar to the use case described above, the difference is on the way the policies are matched and the configuration of the resources to be able to rely on istio’s mTLS between the sidecar and egress:

For Google:

apiVersion: networking.istio.io/v1alpha3
kind: ServiceEntry
metadata:
  name: google
spec:
  hosts:
  - developers.google.com
  location: MESH_EXTERNAL
  ports:
  - number: 443
    name: google-tls
    protocol: TLS
  resolution: DNS
---
apiVersion: networking.istio.io/v1alpha3
kind: Gateway
metadata:
  name: istio-google-egressgateway
spec:
  selector:
    istio: egressgateway # use Istio default gateway implementation
  servers:
  - port:
      number: 443
      name: https-port-for-tls-origination-google
      protocol: TLS
    hosts:
    - developers.google.com
    tls:
      mode: ISTIO_MUTUAL
---
# Routes internal outbound traffic to the egress gateway using Istio's mTLS
apiVersion: networking.istio.io/v1alpha3
kind: DestinationRule
metadata:
  name: egressgateway-for-google
spec:
  host: istio-egressgateway.istio-system.svc.cluster.local
  subsets:
  - name: google
    trafficPolicy:
      portLevelSettings:
      - port:
          number: 443
        tls:
          mode: ISTIO_MUTUAL
          sni: developers.google.com
---
apiVersion: networking.istio.io/v1alpha3
kind: VirtualService
metadata:
  name: direct-google-through-egress-gateway
spec:
  hosts:
  - developers.google.com
  gateways:
  - istio-google-egressgateway
  - mesh
  # route HTTP traffic to developers.google.com through the egress gateway for the entire mesh
  tls:
  - match:
    - gateways:
      - mesh # apply to sidecars in the mesh
      port: 443
      sniHosts:
      - developers.google.com
    route:
    - destination:
        host: istio-egressgateway.istio-system.svc.cluster.local
        subset: google
        port:
          number: 443
      weight: 100
  # at the egress gateway, route developers.google.com to the real destination outside the mesh
  tcp:
  - match:
    - gateways:
      - istio-google-egressgateway
      port: 443
    route:
    - destination:
        host: developers.google.com
        port:
          number: 443
      weight: 100
---

For Yahoo:

apiVersion: networking.istio.io/v1alpha3
kind: ServiceEntry
metadata:
  name: yahoo
spec:
  hosts:
  - developer.yahoo.com
  location: MESH_EXTERNAL
  ports:
  - number: 443
    name: yahoo-tls
    protocol: TLS
  resolution: DNS
---
apiVersion: networking.istio.io/v1alpha3
kind: Gateway
metadata:
  name: istio-yahoo-egressgateway
spec:
  selector:
    istio: egressgateway # use Istio default gateway implementation
  servers:
  - port:
      number: 443
      name: https-port-for-tls-origination-yahoo
      protocol: TLS
    hosts:
    - developer.yahoo.com
    tls:
      mode: ISTIO_MUTUAL
---
# Routes internal outbound traffic to the egress gateway using Istio's mTLS
apiVersion: networking.istio.io/v1alpha3
kind: DestinationRule
metadata:
  name: egressgateway-for-yahoo
spec:
  host: istio-egressgateway.istio-system.svc.cluster.local
  subsets:
  - name: yahoo
    trafficPolicy:
      portLevelSettings:
      - port:
          number: 443
        tls:
          mode: ISTIO_MUTUAL
          sni: developer.yahoo.com
---
apiVersion: networking.istio.io/v1alpha3
kind: VirtualService
metadata:
  name: direct-yahoo-through-egress-gateway
spec:
  hosts:
  - developer.yahoo.com
  gateways:
  - istio-yahoo-egressgateway
  - mesh
  # route HTTP traffic to developers.google.com through the egress gateway for the entire mesh
  tls:
  - match:
    - gateways:
      - mesh # apply to sidecars in the mesh
      port: 443
      sniHosts:
      - developer.yahoo.com
    route:
    - destination:
        host: istio-egressgateway.istio-system.svc.cluster.local
        subset: yahoo
        port:
          number: 443
      weight: 100
  # at the egress gateway, route developer.yahoo.com to the real destination outside the mesh
  tcp:
  - match:
    - gateways:
      - istio-yahoo-egressgateway
      port: 443
    route:
    - destination:
        host: developer.yahoo.com
        port:
          number: 443
      weight: 100
---

And the policies:

For the sleep-yahoo svc SA principal on the otherns ns to block outbound traffic to google matching the sni host:

apiVersion: security.istio.io/v1beta1
kind: AuthorizationPolicy
metadata:
  name: external-deny-developers-google-com
  # No ns means that applies to all ns in a mesh
spec:
  # allow-list for the identities that can call the host
  action: DENY
  rules:
  - from:
    - source:
        principals: ["cluster.local/ns/otherns/sa/sleep-yahoo"]
    when:
    - key: connection.sni
      values:
      - developers.google.com

For the sleep-google service account principal in the otherns namespace to block outbound traffic to yahoo matching the sni:

apiVersion: security.istio.io/v1beta1
kind: AuthorizationPolicy
metadata:
  name: external-deny-developer-yahoo-com
  # No ns means that applies to all ns in a mesh
spec:
  # allow-list for the identities that can call the host
  action: DENY
  rules:
  - from:
    - source:
        principals: ["cluster.local/ns/otherns/sa/sleep-google"]
    when:
    - key: connection.sni
      values:
      - developer.yahoo.com

The key is a critical component when doing TLS origination. Setting the key to sni prevents SSL errors due to a mismatch in the certificate’s common name (SAN) and the host name.

Now apply the resources:

kubectl apply -f external-google-https.yaml -n istio-system
kubectl apply -f external-yahoo-https.yaml -n istio-system
kubectl apply -f authz-policy-deny-google-custom.yaml -n istio-system
kubectl apply -f authz-policy-deny-yahoo-custom.yaml -n istio-system

Now testing you should get the following results (make sure only the two previous policies are in place):

kubectl exec $SLEEP_POD1 -it -- curl -I https://developer.yahoo.com
kubectl exec $SLEEP_POD1 -it -- curl -I https://developers.google.com
Both should be 200.

kubectl exec $SLEEP_POD2 -n otherns -it -- curl -I https://developers.google.com
kubectl exec $SLEEP_POD2 -n otherns -it -- curl -I https://developer.yahoo.com
Both should be 200.

kubectl exec $SLEEP_POD_G -n otherns -it -- curl -I https://developers.google.com
kubectl exec $SLEEP_POD_G -n otherns -it -- curl -I https://developer.yahoo.com
The first one being the google pod should be able to access and get a 200, the second one should be blocked.

kubectl exec $SLEEP_POD_Y -n otherns -it -- curl -I https://developers.google.com
kubectl exec $SLEEP_POD_Y -n otherns -it -- curl -I https://developer.yahoo.com
The first one being the yahoo pod should be blocked because is trying to access google, the second one should be 200.


Summary📜

Using the right combination of ServiceEntry, DestinationRule, VirtualService, Gateway and AuthorizationPolicy resources allows us to route internal outbound traffic to our egress gateway instance with fine-grained control over access and TLS encryption.