10 min read

Google’s long experience and success in, protecting itself against cyberattacks plays to our advantage as customers of the Google Cloud Platform (GCP). From years of warding off security threats, Google is well aware of the security implications of the cloud model. Thus, they provide a well-secured structure for their operational activities, data centers, customer data, organizational structure, hiring process, and user support.

Google uses a global scale infrastructure to provide security to build commercial services, such as Gmail, Google search, Google Photos, and enterprise services, such as GCP and gsuite.

This article is an excerpt taken from the book, “Google Cloud Platform for Architects.“, written by Vitthal Srinivasan, Janani Ravi and Et al. In this book, you will learn about Google Cloud Platform (GCP) and how to manage robust, highly available, and dynamic solutions to drive business objective.

This article gives an insight into the security features in Google Cloud Platform, the tools that GCP provides for users benefit, as well as some best practices and design choices for security.

Security features at Google and on the GCP

Let’s start by discussing what we get directly by virtue of using the GCP. These are security protections that we would not be able to engineer for ourselves. Let’s go through some of the many layers of security provided by the GCP.

  • Datacenter physical security: Only a small fraction of Google employees ever get to visit a GCP data center. Those data centers, the zones that we have been talking so much about, probably would seem out of a Bond film to those that did—security lasers, biometric detectors, alarms, cameras, and all of that cloak-and-dagger stuff.
  • Custom hardware and trusted booting: A specific form of security attacks named privileged access attacks are on the rise. These involve malicious code running from the least likely spots that you’d expect, the OS image, hypervisor, or boot loader. There is the only way to really protect against these, which is to design and build every single element in-house. Google has done that, including hardware, a firmware stack, curated OS images, and a hardened hypervisor. Google data centers are populated with thousands of servers connected to a local network. Google selects and validates building components from vendors and designs custom secure server boards and networking devices for server machines. Google has cryptographic signatures on all low-level components, such as BIOS, bootloader, kernel, and base OS, to validate the correct software stack is booting up.
  • Data disposal: The detritus of the persistent disks and other storage devices that we use are also cleaned thoroughly by Google. This data destruction process involves several steps: an authorized individual will wipe the disk clean using a logical wipe. Then, a different authorized individual will inspect the wiped disk. The results of the erasure are stored and logged too. Then, the erased driver is released into inventory for reuse. If the disk was damaged and could not be wiped clean, it is stored securely and not reused, and such devices are periodically destroyed. Each facility where data disposal takes place is audited once a week.
  • Data encryption: By default GCP always encrypts all customer data at rest as well as in motion. This encryption is automatic, and it requires no action on the user’s part. Persistent disks, for instance, are already encrypted using AES-256, and the keys themselves are encrypted with master keys. All these key management and rotation is managed by Google. In addition to this default encryption, a couple of other encryption options exist as well, more on those in the following diagram:

  • Secure service deployment: Google’s security documentation will often refer to secure service deployment, and it is important to understand that in this context, the term service has a specific meaning in the context of security: a service is the application binary that a developer writes and runs on infrastructure. This secure service deployment is based on three attributes:
    • Identity: Each service running on Google infrastructure has an associated service account identity. A service has to submit cryptographic credentials provided to it to prove its identity while making or receiving remote procedure calls (RPC) to other services. Clients use these identities to make sure that they are connecting to an intended server and the server will use to restrict access to data and methods to specific clients.
    • Integrity: Google uses a cryptographic authentication and authorization technique at an application layer to provide strong access control at the abstraction level for interservice communication. Google has an ingress and egress filtering facility at various points in their network to avoid IP spoofing. With this approach, Google is able to maximize their network’s performance and its availability.
    • Isolation: Google has an effective sandbox technique to isolate services running on the same machine. This includes Linux user separation, language and kernel-based sandboxes, and hardware virtualization. Google also secures operation of sensitive services such as cluster orchestration in GKE on exclusively dedicated machines.
  • Secure interservice communication: The term inter-service communication refers to GCP’s resources and services talking to each other. For doing so, the owners of the services have individual whitelists of services which can access them. Using them, the owner of the service can also allow some IAM identities to connect with the services managed by them.Apart from that, Google engineers on the backend who would be responsible to manage the smooth and downtime-free running of the services are also provided special identities to access the services (to manage them, not to modify their user-input data).
    Google encrypts interservice communication by encapsulating application layer protocols in RPS mechanisms to isolate the application layer and to remove any kind of dependency on network security.
  • Using Google Front End: Whenever we want to expose a service using GCP, the TLS certificate management, service registration, and DNS are managed by Google itself. This facility is called the Google Front End (GFE) service. For example, a simple file of Python code can be hosted as an application on App Engine that (application) will have its own IP, DNS name, and so on.
  • In-built DDoS protections: Distributed Denial-of-Service attacks are very well studied, and precautions against such attacks are already built into many GCP services, notably in networking and load balancing. Load balancers can actually be thought of as hardened, bastion hosts that serve as lightning rods to attract attacks, and so are suitably hardened by Google to ensure that they can withstand those attacks. HTTP(S) and SSL proxy load balancers, in particular, can protect your backend instances from several threats, including SYN floods, port exhaustion, and IP fragment floods.
  • Insider risk and intrusion detection: Google constantly monitors activities of all available devices in Google infrastructure for any suspicious activities. To secure employees’ accounts, Google has replaced phishable OTP second factors with U2F, compatible security keys.
    Google also monitors its customer devices that employees use to operate their infrastructure. Google also conducts a periodic check on the status of OS images with security patches on customer devices. Google has a special mechanism to grant access privileges named application-level access management control, which exposes internal applications to only specific users from correctly managed devices and expected network and geographic locations. Google has a very strict and secure way to manage its administrative access privileges. They have a rigorous monitoring process of employee activities and also a predefined limit for administrative accesses for employees.

Google-provided tools and options for security

As we’ve just seen, the platform already does a lot for us, but we still could end up leaving ourselves vulnerable to attack if we don’t go about designing our cloud infrastructure carefully. To begin with, let’s understand a few facilities provided by the platform for our benefit.

  • Data encryption options: We have already discussed Google’s default encryption; this encrypts pretty much everything and requires no user action. So, for instance, all persistent disks are encrypted with AES-256 keys that are automatically created, rotated, and themselves encrypted by Google.
  • In addition to default encryption, there are a couple of other encryption options available to users.
    • Customer-managed encryption keys (CMEK) using Cloud KMS: This option involves a user taking control of the keys that are used, but still storing those keys securely on the GCP, using the key management service. The user is now responsible for managing the keys that are for creating, rotating and destroying them. The only GCP service that currently supports CMEK is BigQuery and is in beta stage for Cloud Storage.
    • Customer-supplied encryption keys (CSEK): Here, the user specifies which keys are to be used, but those keys do not ever leave the user’s premises. To be precise, the keys are sent to Google as a part of API service calls, but Google only uses these keys in memory and never persists them on the cloud. CSEK is supported by two important GCP services: data in cloud storage buckets as well as by persistent disks on GCE VMs. There is an important caveat here though: if you lose your key after having encrypted some GCP data with it, you are entirely out of luck. There will be no way for Google to recover that data.
      • Cloud security scanner: Cloud security scanner is a GCP, provided security scanner for common vulnerabilities. It has long been available for App Engine applications, but is now also available in alpha for Compute Engine VMs. This handy utility will automatically scan and detect the following four common vulnerabilities:
        • Cross-site scripting (XSS)
        • Flash injection
        • Mixed content (HTTP in HTTPS)
        • The use of outdated/insecure libraries
    • Like most security scanners, it automatically crawls an application, follows links, and tries out as many different types of user input and event handlers as possible.

Some security best practices

Here is a list of design choices that you could exercise to cope with security threats such as DDoS attacks:

  • Use hardened bastion hosts such as load balancers (particularly HTTP(S) and SSL proxy load balancers).
  • Make good use of the firewall rules in your VPC network. Ensure that incoming traffic from unknown sources, or on unknown ports, or protocols is not allowed through.
  • Use managed services such as Dataflow and Cloud Functions wherever possible; these are serverless and so have smaller attack vectors.
  • If your application lends itself to App Engine it has several security benefits over GCE or GKE, and it can also be used to autoscale up quickly, damping the impact of a DDOS attack.
  • If you are using GCE VMs, consider the use of API rate limits to ensure that the number of requests to a given VM does not increase in an uncontrolled fashion.
  • Use NAT gateways and avoid public IPs wherever possible to ensure network isolation.
  • Use Google CDN as a way to offload incoming requests for static content. In the event of a storm of incoming user requests, the CDN servers will be on the edge of the network, and traffic into the core infrastructure will be reduced.

Summary

In this article, you learned that the GCP benefits from Google’s long experience countering cyber-threats and security attacks targeted at other Google services, such as Google search, YouTube, and Gmail. There are several built-in security features that already protect users of the GCP from several threats that might not even be recognized as existing in an on-premise world. In addition to these in-built protections, all GCP users have various tools at their disposal to scan for security threats and to protect their data.

To know more in-depth about the Google Cloud Platform (GCP), head over to the book, Google Cloud Platform for Architects.

Read Next

Ansible 2 for automating networking tasks on Google Cloud Platform [Tutorial]

Build Hadoop clusters using Google Cloud Platform [Tutorial]

Machine learning APIs for Google Cloud Platform

Vincy Davis
A born storyteller turned writer!