Security

November 22, 2026

Where do I even start with the last month itself? I've been fixing CVEs at both work and home for so long that it's starting to become terribly annoying. And when it's not CVEs, it's generally packages in Guix failing to build, requiring me to come up with workarounds to support the fact that it may take a very long time for the package to become fixed.

I have thoughts on the state of things, but first I wanted to cover each one and talk about the scale and impact of each one.

PHP2Shell

PHP2Shell as it came known to be is a remote-code exploit where an attack can get elevated access to a root shell on a host. This is typically very bad, as most of the internet is in fact hosted by PHP. Stellar examples would be Wikipedia, Wordpress, and Nextcloud, the most popular self-hoster's Swiss army knife.

This bug as I've seen it is not totally practical to exploit, as it requires a specific amount of timing information and pretty strange php.ini options, but it would be a matter of time before it would become easier to exploit. Most places try not to rely on PHP anymore since it's overall a pretty awful language to work with, and even Meta swapped to C++ years before this ever happened. However, I still went around to make sure I did a sanity check locally to ensure I have no PHP-backed containers, and I am still PHP-free.

Java2Shell

This one impacts me slightly since I (used to) play a Java-based game called Oldschool Runescape. Runescape does process a lot of random data whether it be from internet packets or from plugins, and for Java2Shell to exist because of bad Base-64 decoding, it's not a great look.

My Oldschool Runescape client Runelite runs as part of a Flatpak, so there's not much I could really do until Flatpak images get updated anyways. But, since my membership expired, this one I didn't really need to care about. I deleted the Flatpak images and made another cursory glance over my containers to see if any Java was used.

Java is still used in many places, but this is a bug that is limited to a specific set of CPU architectures at least. With the move to ARM64 or perhaps even RISC-V, I don't think many will worry as much, unless their Java applications are insanely hard to update.

Nginx2Shell

The reverberations of Nginx2shell will be felt by many, as it is often times the go-to HTTP router people lean on. I do not use Nginx, but instead prefer to use Caddy. This vulnerability however had the chance to go nuts, but Kubernetes removed Nginx long before this ever took place.

This is largely something I feel people have been screaming from the rooftops about for close to a decade now. C language is not a golden bullet programmers should be relying on for anymore. It has flaws, and I think the sooner we acknowledge that, the better. The language is dead stuck in time, locked behind ANSI committees for making changes. We should all be considering Go and Rust here on out, because it's pretty clear what we're using currently is not cutting it anymore.

Ruby2ßhell

In a sillier CVE, turns out trying to upper-case a german letter ß ends up in a heap overflow. Whoops! This silly little bug is a result of encodings and ISO format hell, that much I'm sure of.

I have no reason to think I myself am currently exposed to Ruby, but since Ruby is what mostly drives Shopify, I am certain they are not having a great time over there right now.

Suffice to say, there's been far too many new CVEs dropping, and I don't expect it to actually slow down. The people who are responsible for securing these systems are being fired left and right across the industry in favor of our LLM overlords, and those who are still on payroll are not being paid enough to deal with the nonsense. Our internet modern technology stack relies on a literal stack of misshapen blocks a la xkcd 2347.

There's not much more I can say that hasn't been said before. My hot and spicy opinion is that we should focus on:

  • Reproducibility via declarative package manifests
  • Removing C/C++ as the base layer for everything
  • Put Rust into more places; use Go where you can't
  • Create software bills of materials

At Home

The CVEs above do not terribly apply to my current ongoings in fleshing out my homelab, but not a day goes by without another security release of Forgejo impacting my server. Forgejo notably had a pretty rough start with a nasty RCE earlier in the year. Since then, I've been pretty warey of hosting Forgejo at all. I only started hosting it until a few weeks ago.

To avoid a situation where LLMs would absolutely max out my hardware, I am going to be doing two things:

  • Disabling HTTPS cloning directly from my Forgejo
  • Disable downloading of source tarballs
  • Putting Anubis in front of my Forgejo

The first and second options are mainly because it will take me some time to work out how to use Anubis correctly. Those options may not be totally necessary if Anubis works to block all scrapers, which I do not believe to be bullet-proof still. It can stop some, but most likely not all, at least until scrape farms start leveraging virtual desktops way more.

I will need to spend more time tweaking Forgejo as I go along to see what other holes need to be plugged, but I am still a fan of hosting my code online publicly browsable in this format. For any people curious about my work, storing it all locally here is the best way to showcase what I care about to work on.

That being said, the upside of having a personal code forge is good for cases where your main one happens to disappear because the domain was randomly deleted. Kudos to Codeberg for enduring all that.

Bonus: Broken Krita

I said Krita has been broken on Guix, and it has been broken for a while now. It has a lot to do with the swap from Qt5 to Qt6, and other various changes related to KDE's subsystems, but Krita has not successfully since October 8th 2026.

My home configuration treats Krita as a main dependency, I do use Krita to draw, and granted I can use other programs, I would rather not have to modify my home each and every single time Krita's build processes fail on me. Therefore, I need to downgrade it into an inferior.

I already do this with Firefox as a means of avoiding building Firefox from source, which is hell. But now I can do it with Krita, using a functional build until someone on the KDE front wants to actually address one of the most popular painting programs on Linux not building correctly.

(define (pin-package name verno gc)
  (let*
    ((channels
        (list
         (channel
          (name 'guix)
          (url "https://git.guix.gnu.org/guix.git")
          (commit gc))))
     (inferior (inferior-for-channels channels)))
    (car (lookup-inferior-packages inferior name verno))))

(define ^krita
  (pin-package
   "krita" "5.2.16"
   "74f3b2a3d088e21e9a29a432a4e640e200d388c9"
   ))

Bonus II: iptables

I keep running into re-occurring issues with iptables not working. Some days it works fine, other days it doesn't. Upon rebooting my VPN, it both was not functional because I hadn't rebooted it in so long and had done so many deploy runs on it, and also, the iptables system I had in place simply was not good enough.

This commit tells a few different stories.

  • Doing guix deploy is a "soft" update in that it will deploy new packages, but they will not take effect until the system generally reboots. In the time that my VPN went on for the first time until now, everything was fine, but my disk was missing due to it having changed UUIDs because of me formatting the disk with parted. Guix's stage-zero OS wanted one disk matched by a generated UUID, but because I edited the disk, there was a new UUID in place that Guix did not know about. By changing it to look for the label of the Guix installer, it would use the label under /dev/disk/by-label instead. I had to go into the host's web console to reboot into the earliest generation, then do a guix deploy from my local box while it was in its first generation. Fun stuff, but we recovered.
  • The iptables that I was previously operating off was also an incorrect assumption. I never rebooted that often, so the rules never updated each time I ran guix deploy. When I rebooted my system, iptables-restore would complain about an invalid format of the text itself. To resolve it, I went into the system, copied the existing iptables.rules I had created and were stored in the /gnu/store folder and tried applying iptables-restore manually. I found the error, adjusted the config text, and now the VPN is applying the IP packet rules on reboot correctly.
  • The third one is my IP tables rules for my local server had a typo and I never noticed. Having not rebooted the system in a bit, this one fell through the cracks. It also seems like the COMMIT text also needs a new line at the end in order to count as a valid instruction.

It's hard to say when I'll start looking at what other self-hostable applications I'll start looking at, but this wave of AI-discovered CVEs is making it hard to start finding ones to be excited about. For now I'm going to polish up what I have here for now, build out some programs in my spare time, and get ready for the holiday madness that will start next week.