In the previous article in this series we discussed security requirements. When making any product, requirements are a must, and ensuring you have security built into your requirements from the beginning is the first step to ensure your final product will be of high quality. In this article we will discuss the next phase of the system development life cycle: Design.
As you recall, the system development life cycle generally looks like the image below:
When designing software applications, software architects not only need to worry about what the customer has asked for (business requirements), functional requirements (user requirements, scheduling, system requirements), but also non-functional requirements that are often taken for granted, such as usability, quality, and of course, security.
Unfortunately, when we design applications we often forget to think of all the angles, focusing more on ensuring it works, rather than ensuring that it only works the way we have intended. This is where threat modelling comes in, the process of identifying potential threats to your business and application, and ensure that proper mitigations are in place. This article will focus on what concepts we need to consider when designing for security, and in a future article we will discuss threat modelling.
*Secure by design, in software engineering, means that the software has been designed from the ground up to be secure. Malicious practices are taken for granted and care is taken to minimize impact when a security vulnerability is discovered or on invalid user input. — Wikipedia *
A security flaw is an error in the design of the application that allows a user to perform actions they should not be allowed to do. Malicious or damaging actions. This is a flaw, a problem with the design. We use secure design concepts, security project requirements and perform threat modelling in attempts to avoid or minimize opportunities for design flaws.
A security bug is an implementation issue, a problem with the code, that allows a user to use the application in a malicious way. We perform code review, security testing (many types, during different stages of the project), provide secure coding training, and use secure coding concepts and guidelines in order to protect against security bugs.
The later you fix a problem in the SDLC, the more it will cost. An article from Slashdot states that a bug found in requirements may cost $1 to fix, while in design $10, coding $100 and in testing or release $1000. There are many different estimates of cost all over the internet, but instead of using ‘guesstimates’ to try to explain the idea, let me tell you a story.
Imagine you and your spouse have been saving for years and you are having your dream home built for you. It’s almost done, they are putting on the handles for the cupboards, and rolling out the carpets. It’s at this point that you look at your partner and say, “Oh honey, we have seven children, maybe we should have asked for more than one bathroom?”
Adding a bathroom this late in the construction will cost quite a bit, and make your project late, but you know you cannot continue with only 1 bathroom. You speak to the construction company and they explain that you will have to sacrifice a bedroom to add two more bathrooms or make the living room 1/2 the size. It will also mean your family can’t move in for another month. It will cost an arm and a leg.
This is the same situation for software. When you make design changes last minute they aren’t always pretty, they almost always make you miss deadlines, and they are extremely expensive.
The “not enough bathrooms” problem is something that threat modelling would have found. This is something that secure design concepts might have made visible more early on. This problem is the reason that we need to begin security at the start, not the end, of all projects. This is why we need to ‘push left’.
With this in mind, let’s talk about several secure design concepts that should be discussed when designing software applications.
(using multiple layers of security precautions)
The idea of defence in depth is that security should be applied in layers; one level of defence is not necessarily enough. What happens if an attack gets past your Web Application Firewall (WAF)? I certainly hope you have secure code back there. It just doesn’t make sense to only use one precaution if you can use two or more (assuming it’s not “too expensive”).
For instance, if you call the same input sanitization function every time, why not call it for data from the database? Who knows if whoever put it there sanitized it first? Maybe something was missed? Maybe data was dumped in there from a 3rd party? Sanitizing it as it comes out of the database will take fractions of a millisecond. I wouldn’t call that expensive.
(removing unused resources and code)
The smaller your app, your network, or even your country, the less you have to worry about protecting. If you haven’t released that new feature yet, why do you have the code in your app, but the button “hidden”? If you have a secret page, attackers could find it. If you have a ton of your code commented out, why is it still in the final product? If you have virtual machines or other resources on your network, but you aren’t using them, why are they still there (and likely on the internet)? Doing regular “clean up” of your resources, and ensuring you remove commented code, as well removing unused or “secret” features, are all great ways to ensure there are less options for malicious actors to attack.
Giving everyone exactly how much access and control they need to do their jobs, but nothing more, is the concept of least privilege. Why would a software developer need domain admin rights? Why would an administrative assistant need administrative controls on their PC? This is no different for software. If you are using Role Based Access Control (RBAC) to give users different abilities and powers within your application, you wouldn’t give everyone access to everything, would you? Of course not. Because the more people with access, the more risk there is of someone causing a security issue.
This means several things in regard to developing software, and some of it you’re probably not going to like.
Not only does the software itself need to follow the rules of least privilege, but that least privilege must apply to the people creating the software. Software developers are a huge risk to IT security, if one has malicious intent, or has a bad day and acts carelessly, if they are given too much access… The consequences can be severe.
Let’s leave that there for now and continue further into the secure design rabbit hole.
Whenever something fails in your application it must always fail to a known state, preferably it’s original one. Let’s say you’ve run a transaction to transfer money from one account to another, and there’s an error part way through; you certainly wouldn’t want that money to be in limbo. You would want the money returned to the original account, the user given an error that they need to try again, and the system to log whatever happened. You would not want it to fail into an unknown state, uncertain of where the money is, or if it was transferred multiple times, or if it disappeared altogether. Failing safe means rolling back the transaction and starting again, and handling errors gracefully.
(do not write your own)
I’m sure that many of you were just like I me when I was a new software developer: I thought I was the bee’s knees. I was sure that whatever I wrote was THE BEST version ever created. The fastest and definitely the most efficient. But now that I’ve got a few more years under my belt, and perhaps a bit of maturity, I’ve realized that it’s usually best to leave certain things to the experts, and only write custom code when it is truly needed. This means if you are going to perform encryption, input sanitization, output encoding, use keysor connection strings, or anything else that would be considered a security control, you should use the one available to you in your framework or platform.
When you put comments in your code, ensure that you never save passwords, connection strings, or anything else sensitive. This includes your email address, insider-information about your application, and anything else that could allow an attacker a leg up in regards to attacking your application or organization.
Cross Site Request Forgery (CSRF) is a vulnerability first defined by OWASP, where an attacker convinces the victim to click on a link, the link triggers a transaction within an application (let’s say the purchase of a fancy new TV, to be shipped to the attacker), and because the user was already logged into that account (who doesn’t leave their browser open for days on end?), the vulnerable web application completes the transaction (purchase) and the user is none-the-wiser, until the bill arrives and it is already too late.
The best way to defend against this is to ask the user for something that only the user could provide, before every important transaction (purchase, account deactivation, password change, etc.). This could be asking the user to re-enter their password, complete a captcha, or for a secret token that only the user would have. The most common approach is the secret token.
Pro Tip: users hate captchas.
Always use the authorization functionality available to you in your framework. I know we covered this before, but there’s a reason why everyone does it one of the following ways:
Your production data should not be used for testing, development, or any other purpose than what the business intended. This means a masked (anonymized) dataset should be used for all development and testing, and only your ‘real’ data should be in prod.
This means less people will have access to your data; a reduced attack surface. It also means less employees peeking on personal data. Imagine if you have been using a popular messaging platform and you found out that employees were reading your messages, which you thought were private. This would be a violation of your privacy, and most likely also the user agreement. Segregation of production data would eliminate most opportunities for this type of threat.
Threat modelling, in it’s simplest of forms, is a brainstorming session in search of defining all threats that your application, system or product will likely face. Will people try to intercept your data and sell it on the dark web? Would it have any value if they did? What harm could come if it was? How can we protect against this? These are some of the types of questions you may find yourself doing during a session. You would then test your app and review it’s design to ensure you have properly mitigated these threats.
Threat modelling is such a large topic, that it merits it’s own blog post, as mentioned earlier.
I realize that many people will argue with me that “Security Through Obscurity” is not a true defense tactic, but I beg to differ. It should never be your only defence, but if it is one of many, why not? Many companies do not put their code in open repositories in order to make it much more difficult for competing companies to try to replicate their products. Yes, a malicious actor can try to reverse engineer Windows 10, but who has that kind of time?
Is this defense foolproof? Certainly not. Would I put my code for an unreleased and/or highly valuable product in a public GitHub repo? I think not.
In order to make our applications appear professional we should always catch our errors; no one wants to see a stack trace all over the screen. But there are security concerns to be considered as well.
When a stack trace or unhandled error is shown to the user, it gives details to malicious actors as to what technology stack you are running or other information that could potentially help them plan a better attack against you.
Always catch your errors.
We log security issues so that others may have the joy of auditing later… All kidding aside, if important things are not logged, when there is a security investigation, investigators have nothing to work with. And alerting is to ensure people know about problems in a timely manner.
Ensure you log anything important an investigator may need, but be careful not to log any sensitive information such as SIN numbers, passwords, etc.
Label all of your applicable data as sensitive when you design your data formats and ensure the application treats it that way. Design your app with protecting sensitive data in mind.
Up next in part 4 we will discuss secure coding concepts that can be adopted in order to avoid common security bugs (implementation issues).