systemd: pid1 crash fix

[martinetd]

Motivation for this change

Fixes #147103 (pid1 crash)

While backporting the patch I noticed we weren't up to date, so also updated along the way.

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/)
• 21.11 Release Notes (or backporting 21.05 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.

[SuperSandro2000]

Does rebasing the patches actually do anything? patches can be applied if the hunks have offsets.

[martinetd]

Does rebasing the patches actually do anything? patches can be applied if the hunks have offsets.

Other than removing two patches it didn't seem to do anything, no; this was mostly done to make sure it applies (and some apply method fail on fuzz offset so I wasn't sure about this part)
Happy to remove the rebase noise

[martinetd]

I've removed the rebase altogether, it'll probably need further updating when systemd unfreezes after release while this patch will always stay valid (at least until/if it's also backported upstream)

[arianvp]

I've removed the rebase altogether, it'll probably need further updating when systemd unfreezes after release while this patch will always stay valid (at least until/if it's also backported upstream)

I do not think the freeze applies to minor / bugfix bumps.

[flokli]

Like in #146573 (comment), we shouldn't ship these patches downstream, but poke upstream to release it in systemd-stable.

[martinetd]

Like in #146573 (comment), we shouldn't ship these patches downstream, but poke upstream to release it in systemd-stable.

I see. There were other such patches so I didn't think too much about it, I've asked upstream about it in the systemd PR

[flokli]

Like in #146573 (comment), we shouldn't ship these patches downstream, but poke upstream to release it in systemd-stable.

I see. There were other such patches so I didn't think too much about it, I've asked upstream about it in the systemd PR

For curious ones: That's systemd/systemd#21498 (comment).

[Ma27]

Does rebasing the patches actually do anything? patches can be applied if the hunks have offsets.

For the future, I disagree: keeping patches up-to-date reduces the risk of them being unappliable in the future (if the differences around the change become too large).

@flokli I see this is in systemd-stable, but not yet in a release. Do you think it's sufficient to wait for another stable v249 or should we merge this?

(@martinetd I think this should go to staging btw).

[andir]

Does rebasing the patches actually do anything? patches can be applied if the hunks have offsets.

For the future, I disagree: keeping patches up-to-date reduces the risk of them being unappliable in the future (if the differences around the change become too large).

Adding what I wrote on Matrix some time ago:
We should always (at least on major versions) rebase these patches to avoid running into issues with applying them on the next version.

The issue with patch files vs. a git repository is that the patch files have a limited set of context around each change. If that context isn't suffient anymore the patches fail to apply. My workflow for new systemd versions is typically:

• checkout the old systemd version in a Git repostiry
• git am all the patches
• interactive rebase to new git tag
• git format-patch and replace the files in the repo.

This has been going smooth so far.

[flokli]

@flokli I see this is in systemd-stable, but not yet in a release. Do you think it's sufficient to wait for another stable v249 or should we merge this?

Upstream wanted to wait for some other patches, but I haven't seen anything more getting cherry-picked. I poked them for the tag in systemd/systemd#21498 (comment).