Another example of this was log4j: if you were still using the old 1.x log4j versions, you wouldn't have been vulnerable to the log4shell vulnerability, since it was introduced early in the 2.x series. The old 1.x log4j versions had other known vulnerabilities, but only if you were using less common appenders or an uncommon server mode or a built-in GUI log viewer (!); the most common use of log4j (logging into a local file) was not exposed to any of these, and in fact, you could remove the vulnerable classes and still have a functional log4j setup (see for instance https://www.petefreitag.com/blog/log4j-1x-mitigation/ which I just found on a quick web search).
Did log4shell (and a later vulnerability which could only be exploited if you were using Java 9 or later, because it depended on a new method which was introduced on Java 9) lead people to question whether always being on the "latest and greatest" was a good thing? No, AFAIK the opposite happened: people started to push even harder to keep everything on the latest release, "so that when another vulnerability happens, upgrading to a fixed version (which is assumed to be based on the latest release) will be easy".
Lol, this exact thing happened at my last gig. When I first learned of the vulnerability I panicked, until I found out we were so outdated it didn't affect us. We had a sad laugh about it.
> "so that when another vulnerability happens, upgrading to a fixed version (which is assumed to be based on the latest release) will be easy".
I think there is some truth to this motivation though - if you are on an ancient 1.X version and have to jump a major version of two, that almost always causes pain depending on how critical the service or library is. I don't pretend to know the right answer but I always tend to wait several versions before upgrading so any vulnerabilities or fixes can come by the time I get to the upgrade.
Isn't this still true, generally speaking? How many servers are daily compromised due to out of date, exploitable versions of software versus XZ-like sophisticated attack?
Now we're in a position where most projects update so quickly, that you don't really have a choice. If you need to update one component, there's a good chance it will require that many more as dependencies in your project will require an update to be compatible.
The industry as a whole sacrificed stability and some aspects of security for faster advancements and features. Overall I'd say the net benefit is positive, but its times like these that remind us that perhaps we need to slow things down just a little and do a bit of a course correction to bring things into a better balance.
How many security issues spring from outdated packages vs packages updated too hastily?
A newly-introduced security issue tends to have very limited exploitability, because it's valuable, not-yet well understood, and public exploits are yet to be developed.
Compare to that a similar vulnerability in an older package: chances are that everything about it has been learned and is publicly known. Exploits have become a commodity and are now part of every offensive security distro on the planet. If you run that vulnerable version, there's a real risk that a non-targeted campaign will randomly bite you.
Except in the scenario that is this exact case: Supply chain attacks that are developed with the exploit in mind.
For an old vulnerability, however, any unpatched system is a target. So the individual risk for the average unpatched system is still orders of magnitude higher than in the former scenario.
I still think this is the way to go, but it does move some of the problem around in that you have to pick good cleave lines such that people aren’t forced to pull 5 libraries in for 5 features they need. You want one software domain to use two or three and another to use an overlapping set of three.
Everyone being on different versions results in something like a moving target
I highly encourage people to try freeBSD sometime. Give ports a try (although the modern sense is that poudrie is better even if you want custom-built packages). See how nicely everything works. All the system options you need go into rc.conf (almost uniformly). Everything is documented and you can basically operate the system out of the FreeBSD Handbook documentation (it's not at all comparable to the "how to use a window or a menu" level intro stuff the linux provides). You can't do that when everything is furiously churning every release. everything just works, everything is just documented, it's an experience when you're coming from linux.
and that forum post from 2007 on how to tweak a service script is probably still valid, because BSD hasn't had 3 different init systems over that timespan etc.
just like "engineering is knowing how to build a bridge that barely doesn't fall over", engineering here is knowing what not to churn, and fitting your own work and functionality extensions into the existing patterns etc. like it doesn't have to be even "don't make a bigger change than you have to", you just have to present a stable userland and stable kernel interface and stable init/services interface. the fact that linux doesn't present a stable kernel interface is actually fairly sketchy/poor engineering, it doesn't have to be that way, a large subset of kernel interfaces probably should be stable.
Supply chain attacks are a real threat too, but being conservative about upgrading to improve your security posture is like saying "I'm worried someone with the key to my house might rob me, so I'm going to remove the lock from my door."