Systemd stage 1 fsck

[ElvishJerricco]

Description of changes

Until now, systemd-stage-1 has not supported fsck, which is a major problem.

Incidentally, this will also technically help with hibernation. Currently, the unit ordering isn't quite right, so you'll find /sysroot mounting at the same time that systemd-hibernate-resume is starting, which is very dangerous. The problem is that mount units in stage 1 don't get the After=local-fs-pre.target dependency, which is the sync point that systemd-hibernate-resume orders itself before. But systemd-fsck is ordered after local-fs-pre.target, so it works if fsck is working. Realistically, I think the lack of this ordering on the mount units should be considered a bug in systemd, and the patch to fix that bug is like one line, but I haven't opened an issue/PR about that yet.

Targeting staging because the systemd patch that changes how systemd finds fsck had to be modified.

Things done

• Built on platform(s)
  • x86_64-linux
  • aarch64-linux
  • x86_64-darwin
  • aarch64-darwin
• For non-Linux: Is sandbox = true set in nix.conf? (See Nix manual)
• Tested, as applicable:
  • NixOS test(s) (look inside nixos/tests)
  • and/or package tests
  • or, for functions and "core" functionality, tests in lib/tests or pkgs/test
  • made sure NixOS tests are linked to the relevant packages
• Tested compilation of all packages that depend on this change using nix-shell -p nixpkgs-review --run "nixpkgs-review rev HEAD". Note: all changes have to be committed, also see nixpkgs-review usage
• Tested basic functionality of all binary files (usually in ./result/bin/)
• 23.05 Release Notes (or backporting 22.11 Release notes)
  • (Package updates) Added a release notes entry if the change is major or breaking
  • (Module updates) Added a release notes entry if the change is significant
  • (Module addition) Added a release notes entry if adding a new NixOS module
  • (Release notes changes) Ran nixos/doc/manual/md-to-db.sh to update generated release notes
• Fits CONTRIBUTING.md.

[flokli]

Marking as draft because there's a systemd patch file. It has been merged upstream, but I haven't asked them to backport it to stable. (This is also why it's targeting staging)

There's currently no link to the upstream PR, making it very hard to check if upstream has been poked in the meantime or not.

We normally try to not increase our patch count, so please ask upstream first.

[ElvishJerricco]

@flokli Ah, sorry. I've updated the OP with a link to the PR, and I've asked there if it could be backported to systemd-stable

[ElvishJerricco]

(Note that although that PR was a followup to systemd/systemd#25710, it is not required, because NixOS currently doesn't use the cmdline for root and usrt. But systemd-fsck is generated differently for root and usr solely based on the mountpoint path rather than whether they were provided by the cmdline.)

[ElvishJerricco]

Now that v252.5 is in staging, the new patch file that was needed has been backported and is no longer needed.

[lilyinstarlight]

The diff looks reasonable to me, and the new fsck test passes on my system

I've been running this on my machine (only patching systemdStage1 to avoid too many rebuilds) for a few days now and have not noticed issues, but I admittedly don't have any filesystems marked neededForBoot that need fsck

Other systemd-stage-1 tests I ran all still succeeded as well

Thank you for this work!

[Majiir]

@ElvishJerricco Could you share the reason for removing /etc/fstab from the initrd filesystem? I found that this broke one of my custom initrd units, which had a mount <mountpoint> line. I have a workaround, but I'm curious to understand the reason for this change so that I can choose the best workaround in my config.

[ElvishJerricco]

@Majiir /etc/fstab was never the right choice; the systemd tooling wants you to use root= and mount.usr= on the cmdline. The service initrd-parse-etc.service reads /sysroot/etc/fstab, and so does the systemd-fstab-generator, which is what we're overriding with this environment variable.