Some startups see security as a nice-to-have that can be added months or years after launch. The smart ones realize that dependable security from the beginning means solid performance, satisfied customers, and no precious startup dollars wasted on fraud or incidents. F5 Labs decided to peek under the hood of one of these smart startups: Wanderlust Society. This Seattle-based company was created by a team of Amazon veterans looking to reduce the hassle while increasing the enjoyment of travel planning. Wanderlust Society created a web application that wrangles the long tail of personalized recommendations and the online community for travelers looking to take and share highly curated trips.
Before you begin building your architecture, it’s a good idea to have a well-defined idea of what you want. This goes beyond saying “we want to have the best web app ever!” It means clearly laying out the specific priorities. Wanderlust Society thought this through and set the following as the primary goals for their web application:
- Mobile optimized
- Highly available
- Easily scalable
From these goals, risk assessments can be created. To build a security risk model, you use these goals to anticipate potential threats. Just like the importance of expressing well-defined goals, it’s insufficient to just say things like, “we want our site to be secure.” You need to create a risk model that is actionable so it can support designing defenses. Security can mean different things to different organizations, so risks needs to be detailed. The risk model will be used by developers and architects to make tradeoffs. An additional benefit from this risk analysis process is that it helps communicate security to the entire start-up team, both in terms of significance and the details of how specific threats can cause problems. Wanderlust Society did an excellent job of defining these:
- Unauthenticated users should only be able to read or write data and APIs that are explicitly marked publicly available.
- Authenticated users should only be able to see and change their own data.
- Authenticated users should see shared data from other users.
- Authenticated users should not be able to read or write non-user data such as system utilities and application support data.
- Attackers should not be able to access the system by stealing an authenticated user’s credentials.
- Attackers should not be able to steal authenticated users’ credentials in transit.
- Attackers should not be able to steal/scrape Wanderlust Society data.
- Attackers should not be able to intentionally crash, degrade, or modify site functionality.
This list is by no means carved in stone. It can and should be reviewed periodically and updated as conditions changed. It’s a good start, though. You may have noticed that “authenticated user” is called out but not defined. The details are the next step; what matters at this stage is spelling out what needs to be done, not how. Speaking of the next step…
Architect to Meet Goals and Address Risks
Once Wanderlust Society figured out goals and risks, they worked out architecture and security controls. They came up with the following:
Wanderlust Society built their application in the cloud and they also correctly realized that application security in the cloud is their responsibility. So, they had to build and configure the proper tools to lock things down to their specific risks.
First, the application was designed to respond only to HTTPS requests, so all communication is encrypted. Second, the application was partitioned with firewalls and rules locking down traffic in both directions (to reduce attack surface and exfiltration). Databases are in a restricted, non-public subnet and firewalled to a single port. This reduces the risk of attackers stealing data from users.
Access control is the way you ensure only authorized users can access or change data. The first part of access control is authentication. Passwords are a common way to authenticate users, but they are also fragile and a burden to manage properly. Wanderlust Society chose an alternate method and went with Federated Identity. This means their web application pulls from another trusted authentication repository such as a third-party website where a user is already registered. Wanderlust Society chose to federate from Facebook for two reasons. First, most people already have a Facebook account and this way users don’t have to create and manage another account. Second, Facebook’s infrastructure and platform have been out for years and are used by billions daily, so they’ve been proven reliable and secure.
The other half of access control is authorization: controlling which authenticated users can go where within an application. To securely track users, Wanderlust Society used a request/access token system for all service calls. When a user authenticates, he or she is granted a token tied to the originating client device. Because you never want to trust user input, the token is constantly verified at the server side. Another advantage is that tokens have a limited lifespan and are cleared on logout.
As described in the mobile optimized goal, the Wanderlust Society application was designed to be fast. In addition to the app design, they also leveraged cloud technology for speed, including with cached content delivery networks at the edge for all site images as well as frequently used data. This also reduced load on the web servers.
Being Amazon veterans, the app developers are experts at leveraging Amazon Web Services (AWS). The server instances run in Elastic Compute Cloud (EC2) behind load balancers while CloudWatch is used for monitoring and alarming. Multi-availability zones are also deployed.
Wanderlust Society application services are based on the microservices architecture model3 where applications are small and generally focused around a small set of closely related tasks. This allows services to be independently deployable and expanded. Code is hosted in Docker containers within the EC2 instances, which is scalable to meet Wanderlust Society requirements.
Risks and Matching Controls
Here is more detail on how Wanderlust Society chose to mitigate their resulting risk statements with specific controls. Note that some controls can mitigate multiple risks.
|Specific Security Risk||Defense|
|HTTPS||Access control with Federated Identity & request/access token system||Server-side checks, parameterized SQL statements||High availability and monitoring|
|Unauthenticated users should only be able to read or write data and APIs that are explicitly marked publicly available.||Yes|
|Authenticated users should only be able to see and change their own data.||Yes||Yes|
|Authenticated users should see publically shared user data.||Yes|
|Authenticated users should not be able to read or write non-user data such as system utilities and application support data.||Yes|
|Attackers should not be able to access the system by stealing an Authenticated User’s credentials.||Yes||Yes||Yes|
|Attackers should not be able to steal Authenticated User’s credentials in transit.||Yes|
|Attackers should not be able to steal/scrape Wanderlust Society data.||Yes||Yes|
|Attackers should not be able to intentionally crash, degrade, or modify site functionality.||Yes||Yes|
Their ain’t no such thing as a free lunch, so there are always tradeoffs. One big one was using Facebook to federate identity. A minority of people don’t trust Facebook and refuse to use their service, and some people are just not interested in signing up with Facebook. Those potential customers will probably not choose to join for now. Wanderlust Society could have chosen to integrate with Google or another alternate identity service, but Facebook seemed to align with their desired initial userbase. Supporting federated identities allowed Wanderlust Society to push the development work of building their own secure account creation and login functionality to a future time when they have more resources. A worthwhile tradeoff, since building an authentication system from scratch requires expertise and thorough testing.4 The Internet is littered with the wreckage of weakly implemented authentication systems.
The second tradeoff was using the cloud versus an on-premises solution with their own hardware. Here, Wanderlust Society went back to its core mission: building software that helps people travel, not setting up hardware, keeping it running, scaling it, and patching it. For them, this was a no-brainer; they were going with AWS or one of its competitors. The tradeoff was that competitors could have been cheaper than AWS. But for now, they’re happy with their decision.
At this point, Wanderlust Society has foregone web application firewalls. Again, this is another design tradeoff based on cost, but that layer of protection could be easily added later with minimal architectural disruption.
Wanderlust Society is off to a strong start with shrewd practices, including articulating their goals, doing a risk analysis against those goals, and choosing appropriate responses to counter those risks while weighing the tradeoffs.