The DevOps security story is deceptively simple. It’s based on a few fundamental, straight forward ideas and practices:
Smaller Releases are Safer
One of these ideas is that smaller, incremental and more frequent releases are safer and cause less problems than big bang changes. Makes sense.
Smaller releases contain less code changes. Less code means less complexity and fewer bugs. And less risk, because smaller releases are easier to understand, easier to plan for, easier to test, easier to review, and easier to roll back if something goes wrong.
And easier to catch security risks by watching out for changes to high risk areas of code: code that handles sensitive data, or security features or other important plumbing, new APIs, error handling. At Etsy for example, they identify this code in reviews or pen testing or whatever, hash it, and automatically alert the security team when it gets changed, so that they can make sure that the changes are safe.
Changing the code more frequently may also make it harder for the bad guys to understand what you are doing and find vulnerabilities in your system – taking advantage of a temporary “Honeymoon Effect” between the time you change the system and the time that the bad guys figure out how to exploit weaknesses in it.
And changing more often forces you to simplify and automate application deployment, to make it repeatable, reliable, simpler, faster, easier to audit. This is good for change control: you can put more trust in your ability to deploy safely and consistently, you can trace what changes were made, who made them, and when.
And you can deploy application patches quickly if you find a problem.
“…being able to deploy quick is our #1 security feature”
Effective Approaches to Web Application Security, Zane Lackey
Standardized Ops Environment through Infrastructure as Code
DevOps treats “Infrastructure as Code”: infrastructure configurations are defined in code that is written and managed in the same way as application code, and deployed using automated tools like Puppet or Chef instead of by hand. Which means that you always know how your infrastructure is setup and that it is setup consistently (no more Configuration Drift). You can prove what changes were made, who made them, and when.
You can deploy infrastructure changes and patches quickly if you find a problem.
You can test your configuration changes in advance, using the same kinds of automated unit test and integration test suites that Agile developers rely on – including tests for security.
And you can easily setup test environments that match (or come closer to matching) production, which means you can do a more thorough and accurate job of all of your testing.
Automated Continuous Security Testing
DevOps builds on Agile development practices like automated unit/integration testing in Continuous Integration, to include higher level automated system testing in Continuous Delivery/Continuous Deployment.
You can do automated security testing using something like Gauntlt to “be mean to your code” by running canned attacks on the system in a controlled way.
Other ways of injecting security into Devops include:
- Providing developers with immediate feedback on security issues through self-service static analysis: running Static Analysis scans on every check-in, or directly in their IDEs as they are writing code.
- Helping developers to write automated security unit tests and integration tests and adding them to the Continuous testing pipelines.
- Automating checks on Open Source and other third party software dependencies as part of the build or Continuous Integration, using something like OWASP’s Dependency Check to highlight dependencies that have known vulnerabilities.
Fast feedback loops using automated testing means you can catch more security problems – and fix them – earlier.
Operations Checks and Feedback
DevOps extends the idea of feedback loops to developers from testing all the way into production, allowing (and encouraging developers visibility into production metrics and getting developers and ops and security to all monitor the system for anomalies in order to catch performance problems and reliability problems and security problems.
Adding automated asserts and health checks to deployment (and before start/restart) in production to make sure key operational dependencies are met, including security checks: that the configurations correct, ports that should be closed are closed, ports that should be opened are opened, permissions are correct, SSL is setup properly…
Or even killing system processes that don’t conform (or sometimes just to make sure that they failover properly, like they do at Netflix).
People talking to each other and working together to solve problems
And finally DevOps is about people talking together and solving problems together. Not just developers talking to the business/customers. Developers talking to ops, ops talking to developers, and everybody talking to security. Sharing ideas, sharing tools and practices. Bringing ops and security into the loop early. Dev and ops and security working together on planning and on incident response and learning together in Root Cause Analysis sessions and other reviews. Building teams across silos. Building trust.
Making SecDevOps Work
There’s good reasons to be excited by what these people are doing, the path that they are going down. It promises a new, more effective way for developers and security and ops to work together.
But there are some caveats.
Secure DevOps requires strong engineering disciplines and skills. DevOps engineering skills are still in short supply. And so are information security(and especially appsec) skills. People who are good at both DevOps and appsec are a small subset of these small subsets of the talent available.
Outside of configuration management and monitoring, the tooling is limited – you’ll probably have to write a lot of what you need yourself (which leads quickly back to the skills problem).
A lot more work needs to be done to make this apply to regulated environments, with enforced separation of duties and where regulators think of Agile as “the A Word” (so you can imagine what they think of developers pushing out changes to production in Continuous Deployment, even if they are using automated tools to do it). A small number of people are exploring these problems in a Google discussion group on DevOps for managers and auditors in regulated industries, but so far there are more people asking questions than offering answers.
And getting dev and ops and security working together and collaborating across development, ops and security might take an extreme makeover of your organization’s structure and culture.
Secure DevOps practices and ideas aren’t enough by themselves to make a system secure. You still need all of the fundamentals in place. Even if they are releasing software incrementally and running lots of automated tests, developers still need to understand software security and design security in and follow good software engineering practices. Whether they are using “Infrastructure as Code” or not, Ops still has to design and engineer the datacenter and the network and the rest of the infrastructure to be safe and reliable, and run things in a secure and responsible way. And security still needs to train everyone and followup on what they are doing, run their scans and pen tests and audits to make sure that all of this is being done right.
Secure DevOps is not as simple as it looks. It needs disciplined secure development and secure ops fundamentals, and good tools and rare skills and a high level of organizational agility and a culture of trust and collaboration. Which is why only a small number of organizations are doing this today. It’s not a short term answer for most organizations. But it does show a way for ops and security to keep up with the high speed of Agile development, and to become more agile, and hopefully more effective, themselves.