!PSmBFWNKoXmlQBzUQf:helsinki-systems.de

Stage 1 systemd

83 Members
systemd in NixOs's stage 1, replacing the current bash tooling https://github.com/NixOS/nixpkgs/projects/5127 Servers

Load older messages


SenderMessageTime
10 Mar 2023
@lily:lily.flowers@lily:lily.flowers
In reply to @elvishjerricco:matrix.org
That's not good
Should we try to strip that out or just make our own initrd os-release attrset with less info to begin with?
13:41:45
@janne.hess:helsinki-systems.de@janne.hess:helsinki-systems.de Lily Foster: added a MIT license and a README with the appropriate attribution 13:59:25
@janne.hess:helsinki-systems.de@janne.hess:helsinki-systems.deFeel free to open a PR that mentions the package name in the README once it's upstreamed ;)14:00:24
@elvishjerricco:matrix.org@elvishjerricco:matrix.org
In reply to @lily:lily.flowers
Should we try to strip that out or just make our own initrd os-release attrset with less info to begin with?
Probably just separate common, stage 2, and stage 1. I'm going to need another variant for ukify since that seems to require an os-release as well
16:15:44
@k900:0upti.meK900One thing I wanted to do for bootis but didn't get around to doing is building UKIs with Nix 16:16:36
@k900:0upti.meK900That would be really nice16:16:51
@elvishjerricco:matrix.org@elvishjerricco:matrix.orgIt's not hard. I've got a little script for it somewhere16:16:53
@elvishjerricco:matrix.org@elvishjerricco:matrix.org
  addSections = pkgs.writeShellScriptBin "add-sections" ''
    set -euo pipefail
    stub="$1"
    image="$2"

    stub_line=$(${pkgs.binutils}/bin/objdump -h "$stub" | tail -2 | head -1)
    stub_size=0x$(echo "$stub_line" | awk '{print $3}')
    stub_offs=0x$(echo "$stub_line" | awk '{print $4}')
    next_offs=$((stub_size + stub_offs))
    args=()

    while read sectionName contentsFile; do
      contentsFile="$(readlink -f "$contentsFile")"
      args+=(--add-section "$sectionName"="$contentsFile")
      args+=(--change-section-vma "$sectionName"=$(printf 0x%x $next_offs))
      next_offs=$((next_offs + $(stat -c%s "$contentsFile")))
    done

    set -x
    exec ${pkgs.binutils}/bin/objcopy "''${args[@]}" "$stub" "$image"
  '';
  stub = pkgs.runCommand "stub" {nativeBuildInputs = [addSections];} ''
    mkdir $out
    add-sections ${nixosConfig}/systemd/lib/systemd/boot/efi/linuxx64.efi.stub $out/foo-unsigned.efi <<EOF
      .osrel ${osRelease}
      .linux ${nixosConfig}/kernel
      .initrd ${nixosConfig}/initrd${lib.optionalString kernelBoot "\n.cmdline ${cmdline}"}
    EOF
  '';
16:18:36
@elvishjerricco:matrix.org@elvishjerricco:matrix.org *
  addSections = pkgs.writeShellScriptBin "add-sections" ''
    set -euo pipefail
    stub="$1"
    image="$2"

    stub_line=$(${pkgs.binutils}/bin/objdump -h "$stub" | tail -2 | head -1)
    stub_size=0x$(echo "$stub_line" | awk '{print $3}')
    stub_offs=0x$(echo "$stub_line" | awk '{print $4}')
    next_offs=$((stub_size + stub_offs))
    args=()

    while read sectionName contentsFile; do
      contentsFile="$(readlink -f "$contentsFile")"
      args+=(--add-section "$sectionName"="$contentsFile")
      args+=(--change-section-vma "$sectionName"=$(printf 0x%x $next_offs))
      next_offs=$((next_offs + $(stat -c%s "$contentsFile")))
    done

    set -x
    exec ${pkgs.binutils}/bin/objcopy "''${args[@]}" "$stub" "$image"
  '';
  stub = pkgs.runCommand "stub" {nativeBuildInputs = [addSections];} ''
    mkdir $out
    add-sections ${nixosConfig}/systemd/lib/systemd/boot/efi/linuxx64.efi.stub $out/foo-unsigned.efi <<EOF
      .osrel ${osRelease}
      .linux ${nixosConfig}/kernel
      .initrd ${nixosConfig}/initrd
    EOF
  '';
16:20:44
@elvishjerricco:matrix.org@elvishjerricco:matrix.org But also systemd now has theukify tool to do it, which can easily be used in a nix derivation. It just requires an os-release for whatever reason 16:22:00
@k900:0upti.meK900I know it's not hard, I just wanted to get it into nixpkgs16:25:24
@k900:0upti.meK900And probably wire it up as a bootspec extension 16:25:34
@gdamjan:spodeli.org@gdamjan:spodeli.orgyou can create a custom os-release right16:25:41
@elvishjerricco:matrix.org@elvishjerricco:matrix.org gdamjan: Yea I'm just annoyed by it :P 16:25:58
@gdamjan:spodeli.org@gdamjan:spodeli.orgit's used to identify the image :D16:26:17
@elvishjerricco:matrix.org@elvishjerricco:matrix.orgIs there a mapping anywhere in nixpkgs from system architecture to uefi architecture name? e.g. x86_64 -> x6416:31:09
@elvishjerricco:matrix.org@elvishjerricco:matrix.org Oh, derp. xyzPlatform.efiArch 16:32:46
@lily:lily.flowers@lily:lily.flowersDoes anyone want tmpfiles working in initrd? It was a pretty simple patch to get working, but I ended up deciding I didn't need it. So I'll PR it if someone wants it, but probably won't otherwise18:29:17
@lily:lily.flowers@lily:lily.flowershttps://github.com/lilyinstarlight/nixpkgs/commit/d456921f5c93f963bb5491e671b92e20382d3e5b18:29:18
@elvishjerricco:matrix.org@elvishjerricco:matrix.org

Lily Foster: Possibly? For some reason there was an issue in the networkd PR when that was enabled: https://github.com/NixOS/nixpkgs/pull/169116/files#diff-68fef76b66b99ee77baba62969a434c1f912c2fcf8765bfc884b7aa5b0e70d35R71-R72

It was just a unit ordering issue, so it had nothing to do with its lack of implementation. But I can't remember what it was, considering it's been 9 months since I wrote that comment :P

21:07:33
@elvishjerricco:matrix.org@elvishjerricco:matrix.orgIt's possible that issue only existed back when I was trying to reuse the openssh module verbatim in stage 1, rather than doing the dead simple thing like the PR does now21:08:14
@lily:lily.flowers@lily:lily.flowers
In reply to @elvishjerricco:matrix.org
It's possible that issue only existed back when I was trying to reuse the openssh module verbatim in stage 1, rather than doing the dead simple thing like the PR does now
Yeah as a problem that sounds suspicious. I can probably test with it added back if you need to see how it behaves now
21:24:38
@elvishjerricco:matrix.org@elvishjerricco:matrix.org Lily Foster: Feel free. Though fwiw I kinda doubt tmpfiles would be useful for us in stage 1. 21:26:02
@lily:lily.flowers@lily:lily.flowers
In reply to @elvishjerricco:matrix.org
Lily Foster: Feel free. Though fwiw I kinda doubt tmpfiles would be useful for us in stage 1.
Yeah, me too. Which is why I wasn't gonna PR it unless someone expressed an interest to have tmpfiles in initrd
21:26:32
@lily:lily.flowers@lily:lily.flowers(it's a small patch but it also really doesn't do much good -- though I feel we probably should leave it available to work, even if we don't actually add it to any targets or emit tmpfiles.d stuff by default)21:27:13
@lily:lily.flowers@lily:lily.flowers(so to clarify my stance -- I'd rather not remove the unit if we don't have to, but I also don't care to implement stuff in the tmpfiles/initrd modules for it)21:28:42
@elvishjerricco:matrix.org@elvishjerricco:matrix.orgI'd say let's leave it out. It's just an unnecessary thing that can still cause stuff to break for no reason21:29:15
@lily:lily.flowers@lily:lily.flowersThe unit existing really should cause anything to break (which is why that sounds suspicious that it did -- It's rarely pulled in by anything as far as I remember). But the amount I care is not much, so whichever you're more comfortable with is fine21:30:48
@elvishjerricco:matrix.org@elvishjerricco:matrix.orgWell the existing unit also doesn't do anything :P If we're going to keep it, we should have an implementation. And if we have an implementation, that's logic we have to maintain even if no one's using it21:32:29
11 Mar 2023
@jaen:matrix.orgjaenHi, I've asked previously for help with getting systemd-stage1 to work with my LUKS setup, but didn't get any responses, so I've put it aside for a while. I'm trying again now and I think the only issue I have is getting my unit dependencies wrong, as doing the same steps manually on a running system properly shows Plymouth password prompt and decrypts what it should decrypt.11:52:42

Show newer messages


Back to Room ListRoom Version: 6