Skip to content

Latest commit

 

History

History
274 lines (199 loc) · 11.5 KB

README.md

File metadata and controls

274 lines (199 loc) · 11.5 KB
Raw

Lab 4: Getting SVIDS with SPIFFE-Helper

Prerequisites

  • An AMD, Intel, or Arm 64-bit Linux environment.
  • Familiarity with docker and kubectl command-line interface (CLI) commands is helpful.

Introduction

Coastal Containers Ltd. is a leading container shipping company with a legacy manifest file server system. As part of their digital transformation journey, they recognized the need to improve the security of their system. To address this, they are adopting a Zero Trust security model, leveraging SPIFFE/SPIRE. This approach allows them to identify and authenticate workloads in heterogeneous environments securely.

The challenge? Securing their applications using X.509 SVIDs without a large scale refactoring effort. This lab focuses on using the spiffe-helper utility to retrieve and manage SVIDs on behalf of the applications without having to refactor the applications.

Learning Objectives

  • Understand the role and utility of spiffe-helper.
  • Learn how to retrieve and rotate SVIDs for applications that are not SPIFFE aware.

This will make issuing commands easier in the following steps of this exercise, and will reduce the possibility of reference errors.

Step-by-Step Instructions

Step 1: Provision Infrastructure

To set sail, spin up your demo Kubernetes cluster using Kind by issuing the following make command while in the root lab directory:

make cluster-up

Step 2: Deploying the SPIRE Setup

Initialize the SPIRE setup on your Kubernetes cluster and wait for the SPIRE agent to be up and running:

make deploy-spire spire-wait-for-agent

The SPIRE configuration for the server and the agent are identical with the addition of the SPIFFE CSI Driver.

From the SPIFFE Workload Endpoint the workload API must be served over gRPC, and should prefer Unix Domain Socket transport. The SPIRE agents are deployed as a DaemonSet, so there will be a single instance on every Node in the cluster, and expose the Workload API over a Unix Domain Socket using a hostPath volume i.e. mounting a directory from the host Node's filesystem. For workloads to be able to access the Workload API, they would require permissions to mount hostPath volumes which presents a number of security risks.

To address this problem, the SPIFFE project has created a Container Storage Interface (CSI) driver for Kubernetes that facilitates injection of the SPIFFE Workload API. This is done by mounting a directory containing the Workload API Unix domain socket as an ephemeral inline volume into your workload pods.

This means that the SPIRE agent requires the necessary permissions to mount hostPath volumes, and therefore deploying these would normally be done as part of platform provisioning. The workloads however, do not need these elevated permissions.

Typically, the CSI driver will share the directory hosting the Workload API with a read-only bind mount into the container at the requested target path. This begs the question of what an ephemeral inline volume is, and why is it useful. Traditionally, volumes provided by external storage drivers in K8s are persistent by nature. This means that they operate on a completely independent lifecycle to the pods (e.g., persisting even if a pod is killed). In use cases where data volumes are required to be tied directly to the lifecycle of a pod, they need to be created and deleted with the pod (aka ephemeral). To achieve this, these volumes are defined as part of the pod spec itself (aka inline).

Circling back to the spiffe-csi-driver, this means that when the pod is destroyed, the driver is invoked and removes the bind mount.

Step 3: Create Initial Node Registration Entry

Using your SPIRE tools, create the initial node registration entry. This step ensures your SPIRE server recognizes nodes in your cluster.

kubectl exec -n spire spire-server-0 -- \
    /opt/spire/bin/spire-server entry create \
    -spiffeID spiffe://coastal-containers.example/agent/spire-agent \
    -selector k8s_psat:cluster:kind-kind \
    -selector k8s_psat:agent_ns:spire \
    -selector k8s_psat:agent_sa:spire-agent \
    -node

If the command is successful, you should see an SVID displayed in your terminal.

Step 4: Create Registration Entries for Server and Client Workloads

To identify and authenticate your workloads, exec into the running spire-server and create registration entries.

First, create one for the manifest server:

kubectl exec -n spire spire-server-0 -- \
    /opt/spire/bin/spire-server entry create \
    -spiffeID spiffe://coastal-containers.example/manifest/workload/server \
    -parentID spiffe://coastal-containers.example/agent/spire-agent \
    -selector k8s:ns:default \
    -selector k8s:sa:server

Now do the same for the manifest client:

kubectl exec -n spire spire-server-0 -- \
    /opt/spire/bin/spire-server entry create \
    -spiffeID spiffe://coastal-containers.example/manifest/workload/client \
    -parentID spiffe://coastal-containers.example/agent/spire-agent \
    -selector k8s:ns:default \
    -selector k8s:sa:client

Step 5: Crafting the helper.conf files

Now you are ready to write the configuration files for the manifest-server and manifest-client workloads we will deploy in this lab. First create the files for both of the workloads.

touch {manifest-server,manifest-client}/helper.conf

First edit the server helper.config file to contain the following content.

# server helper.conf

# Socket address of the SPIRE Agent. Update this according to your SPIRE setup.
agent_address = "/spire-agent-socket/agent.sock"

# The command that represents the workload (in this case, Python scripts).
cmd = "python3"
cmd_args = "main.py"

# Signal to send to the Python script to trigger a certificate reload.
renew_signal = "SIGHUP"

# Directory where certificates will be written. This should match the path in your Python scripts.
cert_dir = "/var/run/secrets/svids"

# Names of the files where the SVID, private key, and bundle will be stored.
svid_file_name = "server_cert.pem"
svid_key_file_name = "server_key.pem"
svid_bundle_file_name = "svid_bundle.pem"

As you can see, we are configuring the SPIFFE Helper to know:

  • Where to access the Workload API
    • agent_address = "/spire-agent-socket/agent.sock"
  • How to launch the workload as a subprocess
    • cmd = "python3"
    • cmd_args = "main.py"
  • How to signal the workload to reload it's configuration when the SVIDs are rotated
    • renew_signal = "SIGHUP"
  • Where to write the X.509 materials and which filenames to use
    • cert_dir = "/var/run/secrets/svids"
    • svid_file_name = "server_cert.pem"
    • svid_key_file_name = "server_key.pem"
    • svid_bundle_file_name = "svid_bundle.pem"

Next edit the client helper.config file to contain the following content, tailored to the manifest client.

# client helper.conf

# Socket address of the SPIRE Agent. Update this according to your SPIRE setup.
agent_address = "/spire-agent-socket/agent.sock"

# The command that represents the workload (in this case, Python scripts).
cmd = "python3"
cmd_args = "main.py"

# Signal to send to the Python script to trigger a certificate reload.
renew_signal = "SIGHUP"

# Directory where certificates will be written. This should match the path in your Python scripts.
cert_dir = "/var/run/secrets/svids"

# Names of the files where the SVID, private key, and bundle will be stored.
svid_file_name = "client_cert.pem"
svid_key_file_name = "client_key.pem"
svid_bundle_file_name = "svid_bundle.pem"

Finally, create ConfigMaps for the configuration in the Kind cluster:

kubectl create configmap server-helper-config --from-file=manifest-server/helper.conf
kubectl create configmap client-helper-config --from-file=manifest-client/helper.conf

Step 6: Deploying the Workloads

The manifest server is a simply Python web application that returns a ship's manifest as a json payload over HTTPS. It is not SPIFFE aware, so simply loads its CA Certificate, Server Certificate and Key from the filesystem.

Without changing the Python code, we can craft a Dockerfile to containerise the Python application with the SPIFFE Helper so that the SPIFFE Helper can manage the retrieval and rotation of the X.509 SVIDs and the launching and signalling of the application.

We deploy the manifest server to the Kind cluster mounting the SPIFFE Helper configuration from the ConfigMap we created earlier and the Workload API.

The manifest-client connects to the manifest server using its X.509 SVID, again managed by the SPIFFE Helper, retrieves the manifest, and outputs it to stdout. Its Dockerfile is almost identity to the manifest server, and it also mounts the SPIFFE Helper configuration from the ConfigMap and the Workload API when deployed.

Both the server and the client are configured for mutual TLS:

context.verify_mode = ssl.CERT_REQUIRED
context.load_verify_locations(cafile=TRUST_BUNDLE)

First we need to build the manifest server and client images and load them into the Kind cluster:

make workload-images deploy-workloads

To ensure the server is running smoothly and received it's SVID from the spire-agent, check its logs:

kubectl logs -l app=server

To ensure the client is running smoothly, received it's SVID from the spire-agent, and retrieved the shipping manifest from the server workload, check its logs:

kubectl logs -l app=client

Step 7: Observing the Results

Upon inspecting the client logs, you'll observe the fetched shipping manifest, which confirms the secure communication between the server and client. Dive deeper:

  • SVID Rotation: With spiffe-helper, SVIDs are rotated seamlessly. Monitor the logs to see how SVIDs are renewed over time, ensuring continuous secure communication.

  • Python Integration: Recognize how Python applications, which aren't inherently SPIFFE-aware, can now securely interact within the Kubernetes cluster, thanks to the spiffe-helper.

  • Secure Communication: The client's ability to fetch the shipping manifest from the server underscores the secure communication established between them, facilitated by SPIFFE/SPIRE.

Step 8: Cleanup

To tear down the entire Kind cluster, run:

make cluster-down

Conclusion

⚓ Congrats cap'n! You have successfully provided X.509 SVIDs for non SPIFFE-aware workloads by enlisting the help of the trusty spiffe-helper. You should now have an understanding of how you can integrate SPIFFE into existing applications without having to refactor them to be SPIFFE-aware. With the power of SPIFFE/SPIRE, combined with tools like spiffe-helper, organizations like Coastal Containers Ltd. can secure their existing systems in modern, dynamic environments.