chameleon

Borg or the like with ‘hardcoded’ plaintext/regularly full-disk-encrypted key is acceptable. Someone that has your unencrypted private key sitting on your server has almost certainly already obtained access to the entire set of data you’re backing up, with the backup key itself only meaningfully guarding access to older backups.

The more important thing is to securely keep extra copies in case the server fails. I keep mine in a group in my password manager, one per repo.

(It’s a joke/reference, I guess it’s not 100% known though. My bad.)

I really do hate “I know what I have so you are going to pay whatever number I set” capitalism though, which is what they do here. These registrars figured out a loophole around the redemption grace period and are, from the start, set up to make you lose the domain and then spend significant money on a completely unfair auction where they have the power to plant fake bids, rather than paying the usual static redemption fees that aren’t that excessive.

Heartbreaking: The Worst Capitalist Practice You Know Just Accidentally Picked A Funny Target

You go to the settings and verify it. You don’t have to host anything, just verify that you own the domain via text file or DNS record and choose to set it as your handle. Bluesky’s ATProto has a couple extra layers of indirection and it’s very easy to get a custom handle as a result.

The downside of this setup is that running your own complete network is completely impossible. If you want to follow theonion.com, anyone can find did:plc:a4pqq234yw7fqbddawjo7y35 in the DNS without too much work. That’s the identifier for The Onion’s Bluesky account, and even if they swapped back to .bsky.social, that ID number would stay. But that DID tells you absolutely nothing about where the data is currently hosted.

So how do you figure that out? Well, you register it with https://plc.directory/ which is ran by Bluesky and cannot currently be replaced. There’s fancy cryptography involved that makes it hard for them to spoof data, but they are perfectly capable of simply not giving any data out for any given DID.

I don’t have Obsidian around, but this has been happening elsewhere lately too, almost certainly because of this underlying Electron issue: https://github.com/electron/electron/issues/43819

Unfortunately there’s not much you can do about it. Electron decided to depend on functionality not yet in a released version, and that very interesting choice flows down to everything that updates their Electron on the regular.

Sorry, I’ve had a (self-imposed) busy week, but I have to admit, that also has me rather stumped. As far as I can tell, your second entry should work. If the device is visible in /dev/mapper under a name, it should be able to mount under that name.

The only thing I can think of is that some important module like the ext4 module might be missing somehow? You can get pretty confusing errors when that happens. Dracut is supposed to parse /etc/fstab for everything needed to boot, and maybe that’s not recognizing your root for some reason. dmesg might have some useful info at the end after you try to mount it. If that’s what’s happening, you could try to add add_drivers+=" ext4 " in your dracut.conf and regenerate it (the spaces are important!). But if that’s not it, then I’m probably out of ideas now.

I think you should check your root= line and add a rd.luks.uuid= to make it open it. Dracut will by default open the root FS as /dev/mapper/luks-abcdef... based on the LUKS container UUID. You can get that with cryptsetup luksUUID. /dev/mapper/root is just never going to show up unless you’ve assigned a custom name to that with the barely documented rd.luks.name, and I don’t see that in your setup. The cryptroot and cryptdm parameters aren’t used by Dracut either.

With all of that missing it’s just gonna wait for that /dev/mapper/root to magically show up out of nowhere, without ever trying to open it.

A correct cmdline will probably look something along the lines of root=/dev/mapper/luks-<uuid> modules=sd-mod,usb-storage,ext4 rootfstype=ext4 rootflags=rw,relatime rd.luks.uuid=<uuid> and once opening with passphrase works, you can start to mess with rd.luks.key=/awesome.key (and readd quiet when done debugging, if you want it that way).

ldconfig errors and the missing modules should be fine. musl’s ldconfig is just a bit different but also isn’t required in quite the same way. I don’t think you should need to mess with modules manually. I don’t think you’re using LVM’s userland for your setup, just all the device-mapper kernel modules. Dracut will pull all the necessary bits in for you if you’re setting it up for LUKS.

Dracut may have this functionality already built in via rd.luks.key, so a custom module would really only make sense if you’re trying to do more than that. You can probably get away with just using that if you just want it to work, but if you want to customize stuff:

I suspect your module is running well after the device is already supposed to be cryptsetup opened. The way the default crypt module handles it is by setting up udev configuration in a very early phase, and then having udev request the password a little bit later when it finds the device it’s trying to open, until all devices are ready. It’s a complex mechanism compared to Alpine’s straightforward script, but it’s much more flexible when it comes to ordering of things like RAID/network devices/LUKS/etc.

The result of that is that your code would have to run much earlier. There’s some documentation on how hooks work, and the builtin rd.luks.key / keydev handler runs at cmdline 10. That’s well before your pre-mount, and probably where you’d want to run your code. Based on a cursory inspection of the other code, you could either cryptsetup open it yourself if you use the name it expects (rd.luks.name= cmdline parameter or luks-$luks_container_uuid), or you could use that /tmp/luks.keys mechanism (it’s a dracut-internal thing so you won’t find much documentation, but it lives in crypt-lib.sh, cryptroot-ask.sh and probe-keydev.sh).

As for debugging, the cmdline manpage has a few decent enough options. rd.break=cmdline or similar can force a shell before Dracut goes through a specific phase of hooks. You should be able to manually test doing things similar to your script at that point.

You’d be looking for /usr/share/mkinitfs/initramfs-init . I’ve never customized that myself, but it looks like there’s already some support for a keyfile if you look for KOPT_cryptroot and check that block of code. That looks like it’s mostly set up for a keyfile embedded into the initramfs, but I guess it should be possible to replace that code with something that grabs the keyfile off an USB drive.

I suppose you’d make a copy of it, put it somewhere in /etc or whatever and change the mkinitfs.conf to point to it. init="/etc/whatever/myinitramfs-init" should do the trick since the config file just gets sourced in. That said you’re definitively heading into unknown territory here. It might be easier to just use Dracut or the like instead.

mkinitfs doesn’t support running custom shell hooks. mkinitfs is very, very, very bare-bones custom code and the whole features concept exists only to pull extra files and kernel modules into the initramfs, not for extra logic.

You’d either have to customize the init script itself (not impossible, it’s 1000 lines) and pass -i/set init= in the .conf, or install Dracut/Booster instead (which should “just work” if you apk add them, but I’ve had no need to do so).

All of the cool development-related Nix things like pinning a project to known-good library versions (for regression tests or otherwise) don’t really need you to run NixOS. If you like NixOS then it’s a perfectly usable distro for development work, but all of the powers come from Nix itself, and that can be installed anywhere you feel comfortable with.

The only real pro of running full NixOS is that everything you work on will test a relatively uncommon *nix setup by its nature. Things like developer-only scripts with hardcoded #!/bin/bash shebangs are more likely to break on NixOS than they would on a conventional Linux distro with Nix installed. That’s something potentially worth fixing as it might also hurt the developer experience on *BSD/Mac systems.

That already happens constantly and I’d consider this the consequence of it, rather than the cause. You can only issue so many vetoes before people no longer want to deal with you and would rather move on.

The recent week of Wayland news (including the proposal from a few hours ago to restate NACK policies) is starting to feel like the final attempt to right things before a hard fork of Wayland. I’ve been following wayland-protocols/devel/etc from the outside for a year or two and the vibes have been trending that way for a while.

Digging into the GitLab & related discussions, the main takeaway I got is that FFmpeg’s API supposedly meshes better with what Wine needs to provide to Windows code, simplifying things overall. GST is pretty heavy on asynchronous/background processing, which is normally something I’d consider good for media, but if the API you’re expected to implement is synchronous then I guess it only adds complexity.

Most paid certs aren’t worth much anyway. Payment and delivery info for DV certs isn’t validated by anyone, it’s literally the same concept as Let’s Encrypt. OV and EV are the only ones that theoretically have any value, but nobody is using those ever since they got rid of the URL bar labeling; even Amazon is on DV nowadays.

It depends on if you can feasibly implement compatibility layers for large parts of the “required” but very work-intensive drivers. FreeBSD has the same driver struggles and ended up with LinuxKPI to support AMD/Intel GPUs. I know there’s a whole bunch of toy kernels that implemented compatibility layers for parts of Linux in some fashion too.

It’s a ton of work overall but there’s room to lift enough already existing stuff from Linux to get the ball rolling.

In my experience, most hangs with a message about amdgpu loading on screen are caused by an amdgpu issue of some kind. I’d check to see if amdgpu ends up being loaded correctly via lsmod | grep amdgpu and just a general journalctl -b 0 | grep amdgpu to see if there’s any obvious failures there. Chances are that even if it’s not amdgpu, the real failure is in the journal somewhere.

Could be a wrong setting of hardware.enableRedistributableFirmware (should be true) or the new-ish hardware.amdgpu.initrd.enable (can be either really but either true or false might be more or less reliable on your system).

Gonna add a dissenting “maybe but not really”. YT is really aggressive on this kinda stuff lately and the situation is changing month by month. YT has multiple ways of flagging your IP as potentially problematic and as soon as you get flagged you’re going to end up having to run quite an annoying mess of scripts that may or may not last in the long term. There’s some instructions in a stickied issue on the Invidious repo.

Did a physical-to-virtual-to-physical conversion to upgrade and unbreak a webserver that had been messed up by simultaneously installing packages from Debian and Ubuntu.

It’s a problem in the Secure Boot chain, every system is affected by any vulnerability in any past, present or future bootloader that that system currently trusts. Even if it’s an OS you aren’t using, an attacker could “just” install that vulnerable bootloader.

That said, MS had also been patching their own CVE-2023-24932 / CVE-2024-38058, and disabled the fix for that in this update due to widespread issues with it. I don’t think anyone knows what they’re doing anymore.

My dotfiles aren’t distro-specific because they’re symlinks into a git repo (or tarball) + a homegrown shell script to make them, and that’s about the end of it.

My NixOS configuration is split between must-have CLI tools/nice-to-have CLI tools/hardware-related CLI tools/GUI tools and functions as a suitable reference for non-Nix distros, even having a few comments on what the package names are elsewhere, but installation is ultimately still manual.