Fixing indirect vulnerabilities is one of those complex, tedious and, quite frankly, boring tasks that no one really wants to touch. No one except for Debricked, it seems. Sure, there are lots of ways to do it manually, but can it be done automatically with minimal risk of breaking changes? The Debricked team decided to find out.
A forest full of fragile trees
So, where do you even start?
Firstly, there needs to be a way to fix the vulnerability, which, for indirect dependencies, is no walk in the park. Secondly, it needs to be done in a safe way, or, without anything breaking.
You see, indirect dependencies are introduced deep down the dependency tree and it's very tricky to get to the exact version you want. As Debricked's Head of R&D once put it, "You are turning the knobs by playing around with your direct dependencies and praying to Torvalds that the correct indirect packages are resolved. When Torvalds is in your favour, you have to sacrifice some cloud storage to uncle Bob to make sure the updates don't break your application."
In other words, there really should be an easier, less stressful, way to do it.
In this article, we'll walk you through how solving transitive vulnerabilities can be done manually and, towards the end, show you the Debricked solution, which allows you to do it automatically. If you're really just interested in the solution, I suggest you start scrolling.
Precision surgery on your dependency tree
During the research phase of the graph-database project, or, how Debricked today fixes your open source vulnerabilities at the speed of light, the team stumbled upon some articles explaining how to fix indirect vulnerabilities in NPM.
These are both "Prototype Pollution" vulnerabilities, meaning that arguments are not properly sanitized. Luckily, the maintainers of `minimist` fixed these vulnerabilities in version 1.2.6.
Unfortunately, `mocha` version 7.1.0 resolves `minimist` 0.0.8, which is within the vulnerable range of these vulnerabilities. As suggested by the author of this article, these vulnerabilities can be fixed in a few different ways.
But! What about breaking changes?
The first suggestion is to simply trigger an update of all "indirect dependencies", meaning that we won't actually change the version of `mocha`. To perform this update, simply run `npm update`, delete your `npm.lock` file, and run `npm install`. This regenerates the dependency tree with the latest possible version (according to constraints) of your indirect dependencies. With this method, the risk of breaking changes is very low as you actually don't update any of your root dependencies, just your indirect ones.
Breaking changes occur when the package functionality or interface is not forward compatible, meaning that an update to the package could cause your application to break. Common breaking changes are class/function-removal, change of arguments to a function, or licence-change (watch out for that one!).
But life is not always this easy, and this simple update of the tree will not solve the vulnerability. The problem is that `mkdirp` has actually locked their version of `minimist` to 0.0.8. This means that the contributors of `mkdirp` have come to the conclusion that they are not compatible with newer versions of `minimist`, and forcing the update of `minimist` may introduce breaking changes between `mkdirp` and `minimist`.
So, the million-dollar question is: what version of `mocha` should be used, that in turn trickles down to a safe version of `minimist` without breaking the dependency tree? This is actually a graph problem, which has been described in this article.
What graph algorithm would solve this problem? How NPM resolves dependencies can be a bit complicated, as they are allowed to "split" the dependency tree. This means that they can have multiple versions of one dependency to make sure that we always have a tree that is compatible. To solve the vulnerability, we need to make sure that all instances of `minimist` are safe by updating all roots that can trickle down to `minimist`.
The algorithm used to solve this problem is called "All Max Paths Safe". By walking down the dependency graph and keeping the max versions, all while pruning all other versions of that package in each intersection, we can create an approximate representation of our dependency tree. If the approximation is safe, that means that our real tree will be safe as well!
By performing this algorithm for all potential versions of `mocha`, we find the smallest upgrade to fix this vulnerability. To get the speed we wanted for this algorithm, the team had to build a custom Neo4j procedure, which can handle searching over 100 root versions with a search depth of 30+ in ~150 milliseconds. Speedy, huh?
In this case, we don't have to search very far… as 7.1.1 of `mocha` is safe! This is only a patch update, which indicates that the risk of breaking changes is very low. For less complex cases (like this example), 'npm audit' can help you with their fantastic 'npm audit fix' command.
Don't be ad-hoc, enter the pub-sub-human way of working!
Now, if you got this far (congratulations, very impressive) and thought, "this sounds really complex and like an awful lot of work," don't worry - you're not the only one. Luckily, all this happens completely automatically in the Debricked tool when clicking this little button:
If you're not yet a Debricked user, what are you waiting for? It's free for single devs, smaller teams and open source projects (and if you're a larger organization, fear not. There's a generous free trial). Sign up for free here.