nixos: add per-user groups by default

[bjornfor]

Description of changes

Each user now defaults to primary group <user_name>. Before it was
users for normal users and system users required explicit setting the
group -- because the default was "" (empty) which triggered an assert
saying the group must be set.

This change allows finer grained control over file access, and aligns
NixOS with other distros. Here, the result of useradd -m user1 && ls -ld /home/user1:

Ubuntu 22.04 : drwxr-x--- 2 user1 user1 4096 Oct 28 15:17 /home/user1
Fedora 36    : drwx------ 2 user1 user1 4096 Oct 28 15:17 /home/user1
Debian 11    : drwxr-xr-x 2 user1 user1 4096 Oct 28 15:17 /home/user1
Arch Linux   : drwx------ 2 user1 user1 4096 Oct 28 15:17 /home/user1

The new functionality is gated behind an option called
users.solitaryByDefault, which defaults to true for
system.stateVersion >= 24.05.

Fixes #198296.

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/)
• 22.11 Release Notes (or backporting 22.05 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.

[sternenseemann]

Overall LGTM, not sure if the change is good or bad, but moving with other modern distributions can't be terrible.

[bjornfor]

CC release managers for input: @dasJ @mweinelt

[bjornfor]

Rebased and updated to target NixOS 23.05. Please review!

[qb42]

This looks great, I was just looking for an option to do exactly this.

I do have a couple of questions:

1. Do you plan to change the default homeMode to something like 0750?
2. How will the gid be chosen? I believe in other distros it matches the uid if possible.

[bjornfor]

1. Do you plan to change the default homeMode to something like 0750?

I like it. But in a follow-up PR? Or do you feel it belongs in this one?

2. How will the gid be chosen? I believe in other distros it matches the uid if possible.

I think that's handled by this code:

nixpkgs/nixos/modules/config/update-users-groups.pl

Lines 48 to 60 in 62cace1

	sub allocId {
	my ($used, $prevUsed, $idMin, $idMax, $up, $getid) = @_;
	my $id = $up ? $idMin : $idMax;
	while ($id >= $idMin && $id <= $idMax) {
	if (!$used->{$id} && !$prevUsed->{$id} && !defined &$getid($id)) {
	$used->{$id} = 1;
	return $id;
	}
	$used->{$id} = 1;
	if ($up) { $id++; } else { $id--; }
	}
	die "$0: out of free UIDs or GIDs\n";
	}

Correlating UID and GID would be nice.

[bjornfor]

Rebased to resolve conflicts. Please consider for NixOS 23.05.

[RaitoBezarius]

I'm not too shocked by getting this change for 23.05. But I would rather it early rather than late to smoke out any issue and so we can revert timely.

[RaitoBezarius]

Apologies, it completely slipped from my mind, we missed the window for this change, let's get this merged after the branch-off.

[bjornfor]

About the NIXPKGS_ALLOW_UNSUPPORTED_SYSTEM=1 nix-instantiate --attr darwin.linux-builder --show-trace error.

I found that nixos/modules/profiles/macos-builder.nix disables NixOS modules that reference stateVersion, but disabling nixos/modules/config/users-groups.nix doesn't work because that results in error: The option 'boot.initrd.systemd.groups' does not exist.. It seems users/groups is a bit too low-level of a module to disable.

CC @Gabriella439, author of the darwin builder (ref. #206951). Do you have any ideas/tips?

[Gabriella439]

You'll probably want to cc: @roberth since he added the stateVersion-related stuff to macos-builder.nix. See: #207958

[roberth]

If you could factor lib.versionAtLeast nixosConfig.system.stateVersion "23.11" into an option default, e.g. users.solitaryByDefault or something, we can set a fixed value for that macos-builder.nix, thereby avoiding stateVersion.

[RaitoBezarius]

@bjornfor This is our new round for getting this in, are you able to push this to the finish line and we go for the merge early?

[bjornfor]

@RaitoBezarius: Yes, I'll try to post an update within a week.

[bjornfor]

Should we make users.solitaryByDefault an internal/hidden option, or make it public and document that users can have the latest stateVersion but still get the common "users" group by setting users.solitaryByDefault = false?

[bjornfor]

Should we make users.solitaryByDefault an internal/hidden option, or make it public and document that users can have the latest stateVersion but still get the common "users" group by setting users.solitaryByDefault = false?

I don't think we have opt-out for other NixOS options that change behaviour with stateVersion updates, so I guess it's odd to introduce one now.

[roberth]

I don't think we have opt-out for other NixOS options that change behaviour with stateVersion updates

You've found the problem ;)
We should have those; maybe not everywhere, but a stateVersion-free core system is valuable if you're packaging images.

[bjornfor]

Updated with users.solitaryByDefault option. Please review.

[bjornfor]

Re: CI failure:

Build NixOS manual / nixos (pull_request_target) Failing after 21s details

...seems to be an unrelated Cachix failure?

[bjornfor]

Last rebase/force push was to trigger CI, but I also took the opportunity to improve/correct the release note entry.

[nbdd0121]

This works for statically assigned UIDs, but UID could also be set to null to mean auto-assign upon activation, and that's the default. I don't see any logic in update-users-groups.pl to ensure identical UID and GID are auto-assigned?

[bjornfor]

I don't see any logic in update-users-groups.pl to ensure identical UID and GID are auto-assigned?

Correct, I didn't touch that code. Is that a blocker? You still get per-user groups and everything will work fine, except GID != UID, right?

[nbdd0121]

Users and groups with the same name should have equal uids and gids. We are careful in our static allocation: https://github.com/NixOS/nixpkgs/blob/master/nixos/modules/misc/ids.nix#L671.

Also, currently we dynamic allocate GIDs in 400-999 range which are system IDs, and user UID/GIDs should have 1000-29999.

I think before we land this PR we need to change the allocation script to ensure the constraints are met.

[bjornfor]

Although I agree having GID = UID is preferable, I don't understand why that should be a blocker for this PR. Apparently we're OK with the current dynamic allocation that does not have GID = UID, so why would this PR make a difference? (I guess I should update the release notes entry to not mention the GID = UID property, since it's only valid for static UIDs.)

[nbdd0121]

Current the non-ideal behaviour is only exposed to a services that use both a system user and an identical name groups. This does not affect normal users, nor services that use a fixed UID/GID allocation. But with this change, we're now exposing every single normal user (with state version 24.05+) with this behaviour. It's also more difficult to fix user's file permissions than service states if someone wants to change their GID, since systemd will auto fix ownerships.

Also, I think allocating normal user GIDs to <1000 is a hard blocker.

[nbdd0121]

I wonder if the per-user group switch could happen together with the perlless activation... (background: systemd sysusers have provision to avoid same ID from being used by user/groups with different names)

EDIT: I think currently NixOS sysusers module doesn't look at isSystemUser at all, so it's also broken...

[bjornfor]

In perlless NixOS, what code implements the dynamic UID/GID allocations? Is systemd sysusers used for normal users too?

[nikstur]

See this PR that fixes this behaviour (i.e. let sysusers only create system users): #328926

I've reached the point where the only proper solution to our user management is writing my own application that works like systemd-sysusers but also creates normal users and is able to change some data about users (e.g. passwords).

However, I believe it's worth landing this change regardless as it lays the foundation to build software that does this correctly. Since it's gated behind a flag anyway, I don't see a problem.

Btw: isSystemUser is also fundamentally broken (see #328926 (comment))

[RaHoni]

Are there any updates on this?

[bjornfor]

Are there any updates on this?

I guess I'm waiting for the systemd sysusers thing to stabilize and be the default in NixOS (open-ended thread).

[nyabinary]

Are there any updates on this?

I guess I'm waiting for the systemd sysusers thing to stabilize and be the default in NixOS (open-ended thread).

Userborn will be aiming to be the default instead of sysuser according to #332719