Stand on the shoulders of giantsAbout software composition analysis
In our last post, we reproduced the discovery of a vulnerability in libpng. But that is only a small library, you might say, with a very limited scope and only 556 KiB installed. However, many, many packages depend on it. To see how many packages in the Arch Linux repository depend on libpng we can use pacgraph by Kylee Keen:
More than 14 GiB worth of software depends on libpng! And that is only in the Arch Linux repositories, which is hardly the most popular Linux distribution. Also, the library is the official PNG reference library and is cross-platform, so certainly many other packages in other operating systems depend on it.
Now, back in 2015 when libpng had not yet fixed the low-high palette bug, all the programs and libraries above were also automatically vulnerable to the same issue. Actually this is what happened to Equifax with a vulnerability in Apache Struts. Same with many web services that use OpenSLL with Heartbleed.
If this could happen to such flagships as bash, qt, TeX and xfce, it could happen to your organisation. In fact, this problem is so common that it is part of the 2017 OWASP Top 10: they call it “A9: Using components with known vulnerabilities”.
Given the rapid adoption of Free and Open Source Software (FOSS) by large companies, all of a sudden dependency vulnerability appears to be one hell of a problem. Or rather, as yuppies would like to point out, a “business opportunity”?
Many providers of so-called “software composition analysis” (SCA) (don’t google it) have since appeared in the security scene. Some of them are backed by long-standing companies; most are not. In fact this business has gained such momentum, that it is expected to grow more than 20% each year from now to 2022.
What’s worse, it makes the FOSS, that all these companies owe to, look bad. Yet its adoption is not slowing down and, as we will try to show here, it’s not its fault but rather, the dependent app’s; and also that it’s not a FOSS thing but rather that marketing efforts point towards it.
Today’s applications use on average 30+ libraries, which represent up to 80% of the code. Think of it as your code being only a thin layer upon a building of some tiny, some larger boxes. What SCA does then is look for vulnerabilities inside those boxes with information from external databases, which then become vulnerabilities in your own app:
Instead of going from the alleged solution towards the source of the problem, let’s do it backwards.
FOSS is developed and used by thousands around the world. This can be a double-edged blade: on the one hand, according to “Linus’s Law,” bug-finding and patching should be easier as more eyes are involved.
On the other hand, the lack of centralised guiding makes room for bugs. But then again all kinds of software do.
The difference with proprietary software is that, due to all the restrictions it is less likely that their bugs will become public as soon as they would be on the freer side of things. So expect all vulnerabilities to be zero-day.
So if the source of the problem is not FOSS, what is it? The main reasons why so many companies suffer from A9:
Not knowing used dependencies.
Ignorance of their vulnerabilities.
No continuous scanning for bugs.
Not testing for compatibility.
In essence, it all boils down to a lack of communication between the user and the source of the components.
So what can you do? OWASP recommends the following guidelines to prevent A9:
Trim unnecessary dependencies, features, components etc. That way you have less to check.
Continuously monitor components for updates and vulnerability reports.
Only obtain components from trusted sources.
Make these guidelines into a company policy.
There are specific tools for this purpose: they compare the version of the dependency you are using against both remote repositories (to check for updates) and vulnerability databases (like to find out if any of your dependencies has reported vulnerabilities that have not been fixed yet.
Note that the language-specific tools have to be integrated with the appropriate package manager, like npm or yarn with retire.
A bird’s eye view of how the process should integrate with your development flow is depicted by the following diagram provided by Source:Clear.
We see that every time code is added, the whole system gets scanned for third-party software vulnerabilities and other issues easily identified by Static Analysis when code is not available. This is done by following this procedure:
It is a simple process, really.
Notice that the integration is not fully automatic, and it should not be, since these tools could (and usually do) raise false alarms, so they are reviewed by human security experts.
Internally, the process of scanning for third party software is the same for both proprietary and FOSS software, and it is a simple matter of querying the vulnerabilities databases as described above.
Speaking of integration, you may wonder: What if my app is deployed inside a container? “30% of official images in Docker Hub contain high priority security vulnerabilities”, according to Pentestit. Fortunately, there are tools which go into your container and perform SCA inside of it (and more), like Anchore and Dockerscan.
I know you did search for “Software Composition Analysis” when I suggested you not to. I just know you did. If you didn’t, good for you! Here’s what you’re missing out on:
All of these industry-leading, award-winning, breakthrough-makers, oracles of the tech future want to sell you one thing: static code analysis plus the tools we discussed above.
While static analysis is a valid tool, it’s just a tool. It can scan code and detect vulnerabilities and unhealthy practices, but also encourages late detection and produces a lot of false positives.
You could try hiring such a service, and maybe even try to complement it with dynamic analysis tools like fuzzing and debuggers, but those have their own issues.
But these are no replacement for good old-fashioned human code review. At least at the moment. According to ,
In the future, we might see things like distributed on-demand security testing and machine learning algorithms using support vector machines to try to predict which commits are likely to open vulnerabilities, but in the meantime, stick to the tried-and-true.
with an itch for CS