Test files often represent states that can’t be represented in the library proper. Things like “a tree where node A is a child of B and node B is a child of A”, “the previous instruction repeated x times” where x was never set or there was no previous instruction, or weird combinations of mutually exclusive effects. More often than not, you can’t really generate those using the library itself, as libraries tend to be written to reject those kinds of invalid states (there’s only so much you can do in C but in functional programming land, “make invalid states unrepresentable” is a straight up mantra).

Even if you did manage to do that, using the system under test to generate test data for the system under test is generally not very useful by itself; you’d need some kind of extra protections on top to make sure the actual test files continue to be identical between revisions (like hashing them). Otherwise, a major incompatibility could be easily overlooked. But that also makes it hard to make any kind of valid changes to the library at all. Worse yet, some libraries don’t implement everything needed to generate the test files: even xz is missing pieces, for example there’s an lzip decompressor but not a compressor.

There’s some arguments to be made for separating the test system from the main distribution, but the end result will likely be that nobody runs the testsuite at all. It’s difficult enough to get distros to do it in the first place.

These things are always easy to say in hindsight, but I do believe that a closer review of the build system shenanigans used to install the backdoor would have at least raised some questions.

Nobody noticed it because nobody is reviewing autotools spaghetti and especially not autotools spaghetti that only exists as shipped in a tarball. Minor differences in those files are perfectly normal as the contents of them are copied in from the shared autoconf-archive project, but every distro ships a different version of that, so what any given thing looks like will depend on the maintainer’s computer. And nearly nobody has a good understanding of what any given line in a .m4 file is going to ultimately lead to the execution of regardless, so why bother investigating any differences? The maintainer of Meson has a good take on this.

Shipping tarballs without any form of generated files and having a process to validate release tarballs against the repo would be a good step, but is much easier said than done for a variety of reasons. Same thing can be said for shipping without any form of binary files in the repo, there’s quite high value in integration tests and xz’s README for the test blobs has correctly included this paragraph for 16 years:

Many of the files have been created by hand with a hex editor, thus there is no better “source code” than the files themselves.

For any given tag, GitHub will always have an autogenerated “archive/” link, but the “release/” link is a set of maintainer-uploaded blobs. In this situation, those are the compromised ones. Any distro pulling from an “archive/” link would be unaffected, but I don’t know of any doing that.

The problem with the “archive/” links is that GitHub reserves the right to change them. They’re promising to give notice, but it’s just not a good situation. The “release/” links are only going to change if the maintainer tries something funny, so the distro’s usual mechanisms to check the hashes normally suffice.

NixOS 23.11 is indeed not affected.

You’re looking at the wrong line. NixOS pulled the compromised source tarball just like nearly every other distro, and the build ends up running the backdoor injection script.

It’s just that much like Arch, Gentoo and a lot of other distros, it doesn’t meet the gigantic list of preconditions for it to inject the sshd compromising backdoor. But if it went undetected for longer, it would have met the conditions for the “stage3”/“extension mechanism”.

Sam Jones’s FAQ is by far the best single source, links to other solid sources for more in-depth technical details and also lightly debunks a few things.

The main thing sources online disagree on are which distros are affected. That’s because it’s not a simple yes/no and some distros are taking a nuanced approach in their public communication, while others have chosen the sledgehammer in an attempt to get people to upgrade their systems but keep/kept the nuance in the back room where the audience understood not everything was known yet. Some distros are underselling how vulnerable they were, others are overselling it.

Realistically, I think vendors will be trying to push their crap using this attack as leverage. They did it with Heartbleed, Shellshock and the Log4j issue. Their software won’t/wouldn’t accomplish anything, just like it didn’t with those issues, but they’re sure as hell gonna try to make it seem like it does.

The base runtime pretty much every Flatpak uses includes xz/liblzma, but none of the affected versions are included. You can poke around in a base runtime shell with flatpak run --command=sh org.freedesktop.Platform//23.08 or similar, and check your installed runtimes with flatpak list --runtime.

23.08 is the current latest version used by most apps on Flathub and includes xz 5.4.6. 22.08 is an older version you might also still have installed and includes xz 5.2.12. They’re both pre-backdoor.

It seems there’s an issue open on the freedesktop-sdk repo to revert xz to an even earlier version predating the backdoorer’s significant involvement in xz, which some other distros are also doing out of an abundance of caution.

So, as far as we know: nothing uses the backdoored version, even if it did use that version it wouldn’t be compiled in (since org.freedesktop.Platform isn’t built using Deb or RPM packaging and that’s one of the conditions), even if it was compiled in it would to our current knowledge only affect sshd, the runtime doesn’t include an sshd at all, and they’re still being extra cautious anyway.

One caveat: There is an unstable version of the runtime that does have the backdoored version, but that’s not used anywhere (I don’t believe it’s allowed on Flathub since it entirely defeats the point of it).

Reproducible builds generally work from the published source tarballs, as those tend to be easier to mirror and archive than a Git repository is. The GPG-signed source tarball includes all of the code to build the exploit.

The Git repository does not include the code to build the backdoor (though it does include the actual backdoor itself, the binary “test file”, it’s simply disused).

Verifying that the tarball and Git repository match would be neat, but is not a focus of any existing reproducible build project that I know of. It probably should be, but quite a number of projects have legitimate differences in their tarballs, often pre-compiling things like autotools-based configure scripts and man pages so that you can have a relaxed ./configure && make && make install build without having to hunt down all of the necessary generators.

Won’t help here; this backdoor is entirely reproducible. That’s one of the scary parts.

This is a fun one we’re gonna be hearing about for a while…

It’s fortunate it was discovered before any major releases of non-rolling-release distros were cut, but damn.

Login isn’t necessary, but there is no :latest tag published so you need to pull a version that exists. The current version is at codeberg.org/forgejo/forgejo:1.21.8-0 or at :1.21 if you want one that tracks patch updates (as found in the container registry).

My casual-browsing-only netbook is currently running on a RAID0 setup between the internal eMMC and the microSD card because I think it’s funnier that way. Nothing useful’s stored on there and it’s one nixos-rebuild away from being reinstalled so I don’t mind the inevitable breakage.

sudo mv /etc/default/grub /root/old_etcdefaultgrub to get it out of the way, then sudo dnf reinstall /etc/default/grub to reinstall the package that provides it, giving you a fresh unmodified copy. Should work for practically any config file on Fedora.

This is a shot in the dark, but since the permissions look fine to me, the only other thing that comes to mind is that the SELinux contexts might not have been copied. Fedora is one of the few distros that enables SELinux in enforcing mode right out of the box. That can be very complex to understand if it breaks.

There is a Fedora documentation page about SELinux. The /var/log/audit/audit.log log file should be full of errors relating to your /home if it broke. I believe that stat /home and stat /new_home should display the SELinux context if SELinux is active, and they should be identical.

Also possible I’m totally off the mark, though, it’s just a possibility.

For the port thing, you can set the net.ipv4.ip_unprivileged_port_start sysctl to a lower value like 80 (may need to go lower if you also do email). It also applies to IPv6.

The default of 1024 is for security, but the actual security granted by it is not really that relevant nowadays. It stems from a time where ports < 1024 were used by machines to trust other machines using stuff like rsh & telnet, and before we considered man-in-the-middle attacks to be practical and relevant. Around the start of this millennium, we learned better. Nowadays we use SSH and everything is encrypted & authenticated.

The only particularly relevant risk is that if you lower it enough to also include SSH’s default port 22, some rogue process at startup might make a fake SSH server. That would come along with the scary version of the “host key changed” banner so the risk is not that high. Not very relevant if you’re following proper SSH security practices.

Within 15 days of making the account, to add. If you missed it or decided you’d rather not give them the password to your previous email account (the alt objective for the Gmail-specific thing) you don’t get a second chance.

I get that it’s free but I trust them much less because of the way they handle that.

Note: The HTTP/3 QUIC module is not enabled by default and is considered experimental

Do note that despite not being enabled by default, it is enabled in the official binary packages.

There’s a funny amount of layers to this thing but as far as I’m concerned, if it’s a feature you ship in the default binary packages on your site, that is definitively enough for a CVE even if it’s disabled by default.

Technically always has, ROCm comes with a “backported” amdgpu module and that’s the one they supposedly test/officially validate with. It mostly exists for the ancient kernels shipped with old long-time support distros.

Of course, ROCM being ROCM, nobody is running an officially supported configuration anyway and the thing is never going to work to an suitably acceptable level. This won’t change that, since it’s still built on top of it.

Even worse than that, they need to be able to make an arbitrary container from an arbitrary attacker-provided Dockerfile, or make fairly arbitrary calls to the Docker daemon (in which case you’ve already lost).

They’re rather uninteresting for anyone self-hosting containers as the runc vuln doesn’t offer a way to escape from within an already running container, while the BuildKit vulns all have fairly odd preconditions or require passing untrusted input. Quite the annoyance if you’re running some kind of public cloud or public CI/CD service, though.

DMA-BUF being marked as “unstable” for a decade was a fucking joke. It’s a protocol that’s required to get any kind of meaningful hardware accel going, which nearly every app does nowadays. Within Wayland circles, it’s been understood it’s not going to change for years, as doing so would break nearly every single existing app, yet all kinds of bikeshedding prevented it from being moved to stable.

Hopefully this marks a turning point for many other similarly important protocols stuck in unstable/staging hell too, like pointer constraints and text input. If devs can’t rely on basic functionality to be present and it takes more than say three years to commit to it, it’s time to admit that either the process or the protocol is broken.