fix(docker): declaratively apply groupmod --non-unique for IPA GID conflict

NixOS's group activation uses plain groupmod, which silently skips the
GID change when SSSD exposes the same GID (50010) via the IPA
docker-access group through NSS.  Add an activation script that runs
after the normal 'groups' step and applies groupmod --non-unique so the
local docker group can share GID 50010 with the SSSD-provided IPA group.
If the GID actually changes on a live system the script also restarts
docker.socket + docker.service so the socket is recreated with the new
GID.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

AGENTS.md

@@ -6,12 +6,14 @@ This repository contains flake-based NixOS configurations for Wayne's LAN
 servers and workstation.
 
 The flake exposes NixOS configurations named `<platform>-<buildtype>`
-(platforms: `linode`, `proxmox`, `lxc`; build types: `minimal`, `nix-cache`,
-`server`, `docker`, `gui`, `pxe-boot`, `tailscale-exit-node`, `tor-relay`), generated from `modules/platforms/*`
-and `modules/build-types/*` by the `mkTarget` function in `flake.nix`. Not
-every combination is built — `pxe-boot` has no `linode` variant. See
-`README.md` for the full current target list; treat `flake.nix` as the
-source of truth since this list can drift.
+(platforms: `linode`, `proxmox`, `lxc`, `baremetal`; build types: `minimal`,
+`nix-cache`, `server`, `docker`, `gui`, `pxe-boot`, `tailscale-router`,
+`tor-relay`, `ha-server`), generated from `modules/platforms/*` and
+`modules/build-types/*` by the `mkTarget` function in `flake.nix`. Not every
+combination is built — `pxe-boot` has no `linode` variant, `ha-server` only
+exists on `proxmox`, and `tor-relay` only exists on `lxc`. See `README.md`
+for the full current target list; treat `flake.nix` as the source of truth
+since this list can drift.
 
 Do not deploy, switch, reboot, repartition, format disks, or run destructive
 install commands from this repository unless explicitly asked.

README.md

@@ -10,13 +10,13 @@ pieces composed in `flake.nix`:
 
 - **Platforms** (what it runs on): `linode`, `proxmox`, `lxc`, `baremetal`
 - **Build types** (what it's for): `minimal`, `nix-cache`, `server`, `docker`,
-  `gui`, `pxe-boot`, `tailscale-exit-node`, `tor-relay`
+  `gui`, `pxe-boot`, `tailscale-router`, `tor-relay`, `ha-server`
 
 Not every combination exists — `pxe-boot` has no `linode` variant, since
 PXE/DHCP/TFTP need LAN L2 adjacency that a Linode VPS doesn't have,
-`tor-relay` currently only exists as `lxc-tor-relay`, and `baremetal`
-currently only exists as `baremetal-gui` (the real gui-host hardware). The
-full list:
+`tor-relay` and `ha-server` currently only exist on `lxc`/`proxmox`, and
+`baremetal` currently only exists as `baremetal-gui` (the real gui-host
+hardware). The full list:
 
 | Target | Purpose |
 | --- | --- |
@@ -29,8 +29,9 @@ full list:
 | `linode-gui` / `proxmox-gui` / `lxc-gui` | Cinnamon desktop workstation — previously the flat `nixos` target |
 | `baremetal-gui` | Same Cinnamon desktop workstation, on the real gui-host hardware — ZFS RAID0 root, systemd-boot |
 | `proxmox-pxe-boot` / `lxc-pxe-boot` | HTTP/iPXE boot asset host — previously the flat `pxe-boot` target |
-| `linode-tailscale-exit-node` / `proxmox-tailscale-exit-node` / `lxc-tailscale-exit-node` | Tailscale exit node |
+| `linode-tailscale-router` / `proxmox-tailscale-router` / `lxc-tailscale-router` | Tailscale subnet router + MagicDNS forwarder for the LAN |
 | `lxc-tor-relay` | Tor middle relay |
+| `proxmox-ha-server-1` / `proxmox-ha-server-2` | HA file-server cluster nodes — DRBD + XFS + iSCSI + NFS, managed by Corosync + Pacemaker |
 
 Which variant of a given buildtype is actually deployed isn't tracked
 anywhere in this repo — that's live infrastructure state, not something a
@@ -47,8 +48,7 @@ section for which is which.
 
 Each buildtype's `hosts/<name>/host.nix` carries the per-machine identity
 (hostname, hostId, per-machine secrets, `system.stateVersion`) that must stay
-fixed regardless of which platform it's built for — see
-`flake-target-refactor-spec.md` for the full rationale. Every deployed host
+fixed regardless of which platform it's built for. Every deployed host
 stamps its own active target name into `/etc/flake-target` at build time, so
 `nixos-rebuild switch --flake .#$(cat /etc/flake-target)` always picks up the
 right one even after a platform migration changes the flake attribute name.
@@ -167,7 +167,6 @@ per-boot host key for sops-nix to derive from on ephemeral media) — see
 (`vars/per-machine/<target>/openssh/`, committed and sops-encrypted) for
 their SSH host keys.
 
-This repository's git *history* still contains secrets committed before this
-migration (see `remove-sensetive-info-refactor.md`) — those are being
-scrubbed and rotated separately; don't treat the repo as safe to make public
-until that's finished.
+This repository's git *history* still contains secrets committed before the
+sops-nix migration — those are being scrubbed and rotated separately; don't
+treat the repo as safe to make public until that's finished.

docs/auto-installer.md

@@ -22,7 +22,7 @@ see "LXC hosts" immediately below for why those are different.**
 ## LXC hosts
 
 `lxc-*` targets (`lxc-minimal`, `lxc-nix-cache`, `lxc-server`, `lxc-docker`,
-`lxc-gui`, `lxc-pxe-boot`, `lxc-tailscale-exit-node`, `lxc-tor-relay`) are **not** installed via `auto-install.sh` — the
+`lxc-gui`, `lxc-pxe-boot`, `lxc-tailscale-router`, `lxc-tor-relay`) are **not** installed via `auto-install.sh` — the
 interactive menu deliberately excludes them. Don't try to select one there;
 `nixos-install` would bind-mount `/` onto `/mnt` (LXC containers have no raw
 disk to partition) and then refuse to touch the filesystem it's currently
@@ -133,13 +133,15 @@ Flake outputs:
 ```nix
 nixosConfigurations.installer   # ISO/netboot installer image
 
-packages.x86_64-linux.iso   # installer ISO/netboot image
-packages.x86_64-linux.pxe   # netboot-ipxe + netboot-initrd + netboot-kernel, bundled
+packages.x86_64-linux.iso       # installer ISO/netboot image
+packages.x86_64-linux.pxe       # auto-installer netboot bundle (kernel + initrd + ipxe script)
+packages.x86_64-linux.pxe-minimal  # vanilla NixOS minimal netboot bundle (no installer wiring)
 ```
 
 ```sh
 nix build .#iso
 nix build .#pxe
+nix build .#pxe-minimal
 ```
 
 There's no `nixosConfigurations.proxmox-lxc` (installer-boots-as-an-LXC-

hosts/docker/host.nix

@@ -14,6 +14,11 @@
   };
   boot.zfs.forceImportRoot = false;
 
+  # Only advertise the LAN interface to IPA DNS. Without this, SSSD registers
+  # every Docker bridge (172.x.x.x) as an A record for docker.sweet.home —
+  # the default dyndns.interface = "*" catches them all.
+  security.ipa.dyndns.interface = vars.lxcLanInterface; # eth0
+
   # Preserved from the pre-refactor `docker` target — stateVersion must never
   # be bumped on an already-installed machine.
   system.stateVersion = "25.05";

hosts/nix-cache/host.nix

@@ -20,7 +20,6 @@
   };
 
   services.beszel.agent.environment = {
-    #DOCKER_HOST = "tcp://docker-socket-proxy:2375";
     KEY = "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIFPR9kwtC4TAeTRu46A7+opZsYpxqkRJ+x/ZyB2GWCeG";
   };
 

hosts/server/host.nix

@@ -21,7 +21,6 @@
   };
 
   services.beszel.agent.environment = {
-    #DOCKER_HOST = "tcp://docker-socket-proxy:2375";
     KEY = "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIFPR9kwtC4TAeTRu46A7+opZsYpxqkRJ+x/ZyB2GWCeG";
     EXTRA_FILESYSTEMS = "${vars.storageRoot}/${vars.nfsShares.dockerVolumes.subpath}";
     LOG_LEVEL = "debug";

modules/build-types/ha-server.nix

@@ -15,6 +15,16 @@
 #   under services.beszel.agent.environment.KEY once the hub accepts the
 #   new agents, following the pattern in hosts/server/host.nix.
 { lib, vars, ... }:
+
+let
+  # Generates /etc/exports lines for all nfsShares data entries. Shared
+  # pattern with modules/build-types/server.nix — both export the same
+  # set of shares, differing only in the storage root they serve from.
+  mkNfsExports = storageRoot:
+    lib.concatMapStrings
+      (share: "  ${storageRoot}/${share.subpath} ${vars.lanCidr}${vars.nfsShares.options}\n")
+      (lib.filter builtins.isAttrs (lib.attrValues vars.nfsShares));
+in
 {
   imports = [
     ../ha/pacemaker-stack.nix
@@ -25,16 +35,7 @@
 
   services.nfs.server = {
     enable = true;
-    exports = ''
-      ${vars.haStorageRoot}/${vars.nfsShares.dockerConfig.subpath} ${vars.lanCidr}${vars.nfsShares.options}
-      ${vars.haStorageRoot}/${vars.nfsShares.dockerVolumes.subpath} ${vars.lanCidr}${vars.nfsShares.options}
-      ${vars.haStorageRoot}/${vars.nfsShares.dockerDatabases.subpath} ${vars.lanCidr}${vars.nfsShares.options}
-      ${vars.haStorageRoot}/${vars.nfsShares.nextcloudData.subpath} ${vars.lanCidr}${vars.nfsShares.options}
-      ${vars.haStorageRoot}/${vars.nfsShares.raspiVolumes.subpath} ${vars.lanCidr}${vars.nfsShares.options}
-      ${vars.haStorageRoot}/${vars.nfsShares.proxmoxIsos.subpath} ${vars.lanCidr}${vars.nfsShares.options}
-      ${vars.haStorageRoot}/${vars.nfsShares.proxmoxLxcImages.subpath} ${vars.lanCidr}${vars.nfsShares.options}
-      ${vars.haStorageRoot}/${vars.nfsShares.pxebootImages.subpath} ${vars.lanCidr}${vars.nfsShares.options}
-    '';
+    exports = mkNfsExports vars.haStorageRoot;
   };
 
   # Pacemaker controls nfs-server — prevent systemd from starting it at boot

modules/build-types/server.nix

@@ -13,6 +13,15 @@ let
     lib.concatMap (share: ancestors share.subpath)
       (lib.filter builtins.isAttrs (lib.attrValues vars.nfsShares))
   );
+
+  # Generates /etc/exports lines for all nfsShares data entries (every
+  # attrset value — excludes the bare `options` string). Both server and
+  # ha-server export the same share set from different storage roots, so
+  # this helper is the single source of truth for the export line format.
+  mkNfsExports = storageRoot:
+    lib.concatMapStrings
+      (share: "  ${storageRoot}/${share.subpath} ${vars.lanCidr}${vars.nfsShares.options}\n")
+      (lib.filter builtins.isAttrs (lib.attrValues vars.nfsShares));
 in
 {
   imports = [
@@ -95,16 +104,7 @@ in
 
   services.nfs.server = {
     enable = true;
-    exports = ''
-      ${vars.storageRoot}/${vars.nfsShares.dockerConfig.subpath} ${vars.lanCidr}${vars.nfsShares.options}
-      ${vars.storageRoot}/${vars.nfsShares.dockerVolumes.subpath} ${vars.lanCidr}${vars.nfsShares.options}
-      ${vars.storageRoot}/${vars.nfsShares.dockerDatabases.subpath} ${vars.lanCidr}${vars.nfsShares.options}
-      ${vars.storageRoot}/${vars.nfsShares.nextcloudData.subpath} ${vars.lanCidr}${vars.nfsShares.options}
-      ${vars.storageRoot}/${vars.nfsShares.raspiVolumes.subpath} ${vars.lanCidr}${vars.nfsShares.options}
-      ${vars.storageRoot}/${vars.nfsShares.proxmoxIsos.subpath} ${vars.lanCidr}${vars.nfsShares.options}
-      ${vars.storageRoot}/${vars.nfsShares.proxmoxLxcImages.subpath} ${vars.lanCidr}${vars.nfsShares.options}
-      ${vars.storageRoot}/${vars.nfsShares.pxebootImages.subpath} ${vars.lanCidr}${vars.nfsShares.options}
-    '';
+    exports = mkNfsExports vars.storageRoot;
   };
 
   # mountd (20048) is needed for showmount/NFSv3 mount protocol — without it

modules/common/configuration.nix

@@ -25,13 +25,10 @@ let
   '';
 in
 {
-  imports =
-    [
-      # Include the results of the hardware scan.
-      #      ./hardware-configuration.nix
-      ./set-locale.nix
-      ../ipa/client.nix
-    ];
+  imports = [
+    ./set-locale.nix
+    ../ipa/client.nix
+  ];
 
   # System-wide shell config so all users (including IPA accounts) get the
   # same management aliases as the local nixos user's Home Manager provides.
@@ -42,11 +39,7 @@ in
     };
     interactiveShellInit = buildImageFn;
   };
-  # Use the GRUB 2 boot loader.
-  # boot.loader.grub.enable = true;
-  #boot.loader.grub.device = "/dev/sda"; # or "nodev" for efi only
-
-  networking.networkmanager.enable = true; # Easiest to use and most distros use this by default.
+  networking.networkmanager.enable = true;
 
   # Recommended over the true default (bypasses ZFS's own import safeguards)
   # per the option's own docs; matches hosts/docker/host.nix and

modules/common/preserve-ssh-host-key.nix

@@ -0,0 +1,35 @@
+# Shared activation-script logic to preserve the SSH host key across
+# nixos-rebuild on platforms that embed the key via environment.etc (lxc and
+# proxmox).  When NIXOS_HOST_KEYS_DIR is not set the key is absent from
+# environment.etc, and NixOS's etc activation removes any /etc file not in
+# the new generation — which would destroy the live key and break sops-nix
+# decryption permanently.  These scripts save the key to /run before etc
+# removes it, then restore it afterward.
+#
+# Explicit deps enforce the correct ordering: without them the topological
+# sort places preserveSshHostKey after etc (confirmed live on lxc-tor-relay:
+# position 7 vs etc's position 5), so the key is gone before it can be saved.
+_: {
+  system.activationScripts = {
+    preserveSshHostKey = ''
+      if [ -f /etc/ssh/ssh_host_ed25519_key ]; then
+        cp /etc/ssh/ssh_host_ed25519_key /run/sshd-host-key-preserve.tmp
+        cp /etc/ssh/ssh_host_ed25519_key.pub /run/sshd-host-key-preserve.pub.tmp
+      fi
+    '';
+
+    restoreSshHostKey = {
+      deps = [ "etc" ];
+      text = ''
+        if [ ! -f /etc/ssh/ssh_host_ed25519_key ] && [ -f /run/sshd-host-key-preserve.tmp ]; then
+          install -m 0600 /run/sshd-host-key-preserve.tmp /etc/ssh/ssh_host_ed25519_key
+          install -m 0644 /run/sshd-host-key-preserve.pub.tmp /etc/ssh/ssh_host_ed25519_key.pub
+        fi
+        rm -f /run/sshd-host-key-preserve.tmp /run/sshd-host-key-preserve.pub.tmp
+      '';
+    };
+
+    etc = { deps = [ "preserveSshHostKey" ]; };
+    setupSecrets = { deps = [ "restoreSshHostKey" ]; };
+  };
+}

modules/docker/enable-service.nix

@@ -1,23 +1,45 @@
-{ pkgs, vars, ... }:
+{ lib, pkgs, vars, ... }:
 
+let
+  gid = toString vars.dockerAccessGid;
+in
 {
-  #  virtualisation.docker.enable = true;
   virtualisation.docker = {
     enable = true;
     package = pkgs.docker;
-    #  listenOptions = [
-    #  "unix:///var/run/docker.sock"
-    #  "tcp://0.0.0.0:2375"
-    #];
-
-    #  daemon.settings = {
-    #    metrics-addr = "0.0.0.0:9323";
-    #    experimental = true;
-    #  };
   };
+  # Pin the docker group GID to match the IPA "docker-access" group so that
+  # IPA group membership alone grants access to the Docker socket. Any user
+  # whose supplementary groups (resolved by SSSD from IPA) include GID
+  # vars.dockerAccessGid will pass the socket group-permission check without
+  # any per-host users.groups.docker.members entry.
+  users.groups.docker.gid = lib.mkForce vars.dockerAccessGid;
   users.users.${vars.primaryUser}.extraGroups = [ "docker" ];
   environment.systemPackages = with pkgs; [
     docker-compose
     docker-buildx
   ];
+
+  # NixOS's group activation uses plain `groupmod` without --non-unique.
+  # When SSSD is active it exposes the IPA "docker-access" group at
+  # vars.dockerAccessGid via NSS, so groupmod sees that GID as already in
+  # use and silently skips the change (warning: "not applying GID change").
+  # This script runs after the normal "groups" step and applies the change
+  # with --non-unique (which lets the local docker group share the GID with
+  # the SSSD-provided IPA group). If the GID actually changed it also
+  # restarts docker.socket so the socket is recreated with the new GID.
+  system.activationScripts.docker-group-gid = {
+    deps = [ "groups" ];
+    text = ''
+      current=$(grep "^docker:" /etc/group | cut -d: -f3)
+      if [ "$current" != "${gid}" ]; then
+        ${pkgs.shadow}/bin/groupmod --non-unique -g ${gid} docker
+        if ${pkgs.systemd}/bin/systemctl is-active --quiet docker.socket; then
+          ${pkgs.systemd}/bin/systemctl stop docker.service docker.socket
+          rm -f /var/run/docker.sock
+          ${pkgs.systemd}/bin/systemctl start docker.socket docker.service
+        fi
+      fi
+    '';
+  };
 }

modules/ipa/client.nix

@@ -38,47 +38,122 @@ lib.mkIf enabled {
   networking.domain = lib.mkDefault vars.homeDomain;
   networking.nameservers = lib.mkDefault [ vars.domainControllerIp ];
 
-  security.ipa = {
-    enable = true;
-    domain = vars.homeDomain;
-    inherit realm;
-    server = vars.ipaServer;
-    certificate = caCertPkg;
-    inherit basedn;
-    ipaHostname = fqdn;
-    offlinePasswords = true;
-    cacheCredentials = true;
+  security = {
+    ipa = {
+      enable = true;
+      domain = vars.homeDomain;
+      inherit realm;
+      server = vars.ipaServer;
+      certificate = caCertPkg;
+      inherit basedn;
+      ipaHostname = fqdn;
+      offlinePasswords = true;
+      cacheCredentials = true;
+    };
+
+    # Create the home directory on first login if it doesn't exist yet.
+    # IPA users have no pre-created home on the host; without this sshd
+    # opens a session to a non-existent directory and resets the connection.
+    # lightdm also needs this so the GUI login path can create the home dir
+    # if it was not pre-seeded by the tmpfiles rule above (e.g. on first boot
+    # before SSSD has resolved the user).
+    pam.services = {
+      sshd.makeHomeDir = true;
+      lightdm.makeHomeDir = true;
+
+      # pam_unix returns PAM_AUTHINFO_UNAVAIL without prompting when the local
+      # stub has "!" in shadow (account locked), so PAM_AUTHTOK is never set
+      # and pam_sss's use_first_pass fails with "No authentication token".
+      # Changing to try_first_pass makes pam_sss prompt independently when no
+      # prior module has set the token, restoring IPA password login via
+      # LightDM and su.
+      login.rules.auth.sss.settings = lib.mkForce { try_first_pass = true; };
+      su.rules.auth.sss.settings = lib.mkForce { try_first_pass = true; };
+    };
+
+    # HM with useUserPackages = true (flake.nix) sets users.users.${ipaUser}.packages,
+    # which forces the stub into /etc/passwd. pam_sss.so with the "localusers" flag
+    # (added by NixOS when SSSD is enabled) then skips SSSD for any user it finds in
+    # local /etc/passwd — including this stub — falling through to pam_unix, which has
+    # no password for the stub → sudo auth always fails.
+    #
+    # Fix: NOPASSWD for the IPA user. The IPA user already authenticated to reach a
+    # shell (SSH public key from IPA or Kerberos), so re-prompting via a broken PAM
+    # path is security theater on a single-admin homelab.
+    sudo.extraRules = [{
+      users = [ vars.ipaUser ];
+      commands = [{ command = "ALL"; options = [ "NOPASSWD" ]; }];
+    }];
   };
 
-  # Fetch SSH public keys from IPA so users can log in with the key stored
-  # in their IPA profile rather than needing ~/.ssh/authorized_keys on every
-  # host. sss_ssh_authorizedkeys queries SSSD (which queries IPA LDAP).
-  #
-  # /nix/store is 1775 (group-writable by nixbld). OpenSSH 10.0+ rejects
-  # AuthorizedKeysCommand binaries whose path contains any group-writable
-  # component, silently skipping the command. Copy to /usr/local/bin (all
-  # components root-owned, 755) so the path passes sshd's safety check.
-  systemd.tmpfiles.rules = [
-    "d /usr/local 0755 root root - -"
-    "d /usr/local/bin 0755 root root - -"
-    "C+ /usr/local/bin/sss_ssh_authorizedkeys 0555 root root - ${pkgs.sssd}/bin/sss_ssh_authorizedkeys"
-    # Pre-create the IPA user's home dir so Home Manager activation succeeds
-    # even before their first login. On a fresh system SSSD may not have
-    # resolved the user yet — tmpfiles warns and skips in that case (non-fatal),
-    # and pam_mkhomedir covers the first-login path as a fallback.
-    "d /home/${vars.ipaUser} 0700 ${vars.ipaUser} ${vars.ipaUser} - -"
-  ];
+  systemd = {
+    # Fetch SSH public keys from IPA so users can log in with the key stored
+    # in their IPA profile rather than needing ~/.ssh/authorized_keys on every
+    # host. sss_ssh_authorizedkeys queries SSSD (which queries IPA LDAP).
+    #
+    # /nix/store is 1775 (group-writable by nixbld). OpenSSH 10.0+ rejects
+    # AuthorizedKeysCommand binaries whose path contains any group-writable
+    # component, silently skipping the command. Copy to /usr/local/bin (all
+    # components root-owned, 755) so the path passes sshd's safety check.
+    tmpfiles.rules = [
+      "d /usr/local 0755 root root - -"
+      "d /usr/local/bin 0755 root root - -"
+      "C+ /usr/local/bin/sss_ssh_authorizedkeys 0555 root root - ${pkgs.sssd}/bin/sss_ssh_authorizedkeys"
+      # Pre-create the IPA user's home dir so Home Manager activation succeeds
+      # even before their first login. On a fresh system SSSD may not have
+      # resolved the user yet — tmpfiles warns and skips in that case (non-fatal),
+      # and pam_mkhomedir covers the first-login path as a fallback.
+      "d /home/${vars.ipaUser} 0700 ${vars.ipaUser} ${vars.ipaUser} - -"
+    ];
+
+    # security.ipa enables Kerberos (security.krb5) which causes systemd to
+    # start auth-rpcgss-module.service and rpc-gssd.service for Kerberos NFS
+    # authentication. LXC containers can't load the auth_rpcgss kernel module
+    # and don't have /var/lib/nfs/rpc_pipefs, so both services fail.
+    #
+    # The NixOS IPA module already adds a drop-in for auth-rpcgss-module.service
+    # with ConditionPathExists=/etc/krb5.keytab. We use lib.mkForce to win the
+    # text conflict and add ConditionVirtualization=!container alongside it so
+    # the service is skipped (not failed) in containers that do have a keytab.
+    # Same fix for rpc-gssd.service which also fails in containers.
+    units = lib.mkIf config.boot.isContainer {
+      "auth-rpcgss-module.service" = {
+        overrideStrategy = "asDropinIfExists";
+        text = lib.mkForce ''
+          [Unit]
+          ConditionPathExists=
+          ConditionPathExists=/etc/krb5.keytab
+          ConditionVirtualization=!container
+        '';
+      };
+      # rpc-gssd also has ConditionPathExists from the NixOS IPA module (and an
+      # X-Restart-Triggers store path from systemd.nix). Use mkForce to win;
+      # omit X-Restart-Triggers since this service is skipped in containers anyway.
+      "rpc-gssd.service" = {
+        overrideStrategy = "asDropinIfExists";
+        text = lib.mkForce ''
+          [Unit]
+          ConditionPathExists=
+          ConditionPathExists=/etc/krb5.keytab
+          ConditionVirtualization=!container
+        '';
+      };
+    };
+
+    # home-manager-<user>.service fails on first enrollment because /home/wayne
+    # doesn't exist until the user's first login (pam_mkhomedir creates it then).
+    # ConditionPathExists makes systemd skip the service (exit 0, condition not
+    # met) instead of failing. After first login the dir exists and subsequent
+    # rebuilds activate HM normally.
+    services."home-manager-${vars.ipaUser}".unitConfig.ConditionPathExists =
+      "/home/${vars.ipaUser}";
+  };
 
   services.openssh.extraConfig = ''
     AuthorizedKeysCommand /usr/local/bin/sss_ssh_authorizedkeys %u
     AuthorizedKeysCommandUser nobody
   '';
 
-  # Create the home directory on first login if it doesn't exist yet.
-  # IPA users have no pre-created home on the host; without this sshd
-  # opens a session to a non-existent directory and resets the connection.
-  security.pam.services.sshd.makeHomeDir = true;
-
   # Host keytab: pre-provisioned on the IPA server, sops-encrypted binary.
   # Placed at /etc/krb5.keytab before SSSD starts so the host authenticates
   # to IPA without running ipa-client-install.
@@ -92,44 +167,6 @@ lib.mkIf enabled {
     restartUnits = [ "sssd.service" ];
   };
 
-  # security.ipa enables Kerberos (security.krb5) which causes systemd to
-  # start auth-rpcgss-module.service and rpc-gssd.service for Kerberos NFS
-  # authentication. LXC containers can't load the auth_rpcgss kernel module
-  # and don't have /var/lib/nfs/rpc_pipefs, so both services fail.
-  #
-  # The NixOS IPA module already adds a drop-in for auth-rpcgss-module.service
-  # with ConditionPathExists=/etc/krb5.keytab. We use lib.mkForce to win the
-  # text conflict and add ConditionVirtualization=!container alongside it so
-  # the service is skipped (not failed) in containers that do have a keytab.
-  # Same fix for rpc-gssd.service which also fails in containers.
-  systemd.units = lib.mkIf config.boot.isContainer {
-    "auth-rpcgss-module.service" = {
-      overrideStrategy = "asDropinIfExists";
-      text = lib.mkForce ''
-        [Unit]
-        ConditionPathExists=
-        ConditionPathExists=/etc/krb5.keytab
-        ConditionVirtualization=!container
-      '';
-    };
-    # rpc-gssd also has ConditionPathExists from the NixOS IPA module (and an
-    # X-Restart-Triggers store path from systemd.nix). Use mkForce to win;
-    # omit X-Restart-Triggers since this service is skipped in containers anyway.
-    "rpc-gssd.service" = {
-      overrideStrategy = "asDropinIfExists";
-      text = lib.mkForce ''
-        [Unit]
-        ConditionPathExists=
-        ConditionPathExists=/etc/krb5.keytab
-        ConditionVirtualization=!container
-      '';
-    };
-  };
-
-  # Home Manager config for the IPA primary user, applied on every enrolled
-  # host. Manages what IPA doesn't: dotfiles, user-scoped packages, session
-  # variables. Switch-nix/Test-nix/buildImage are system-wide (configuration.nix)
-  # so they don't need to be repeated here.
   # NixOS requires isNormalUser/isSystemUser + group on any entry in
   # users.users. HM with useUserPackages = true (set in flake.nix) adds a stub
   # entry for each HM user so it can install packages to
@@ -137,35 +174,20 @@ lib.mkIf enabled {
   # With security.ipa setting "passwd: sss files" in nsswitch, SSSD's IPA entry
   # takes priority for NSS lookups — this local stub is only a fallback when
   # SSSD is unreachable (at which point auth fails anyway).
-  # HM with useUserPackages = true (flake.nix) sets users.users.${ipaUser}.packages,
-  # which forces the stub into /etc/passwd. pam_sss.so with the "localusers" flag
-  # (added by NixOS when SSSD is enabled) then skips SSSD for any user it finds in
-  # local /etc/passwd — including this stub — falling through to pam_unix, which has
-  # no password for the stub → sudo auth always fails.
-  #
-  # Fix: NOPASSWD for the IPA user. The IPA user already authenticated to reach a
-  # shell (SSH public key from IPA or Kerberos), so re-prompting via a broken PAM
-  # path is security theater on a single-admin homelab.
   users.users.${vars.ipaUser} = {
     isNormalUser = true;
     group = "users";
     extraGroups = [ "wheel" ];
     createHome = false;
+    # "!" is not a password hash — it is the standard "account locked" marker.
+    # It cannot authenticate anyone locally. It exists solely so NixOS generates
+    # a shadow entry for this stub user; without one pam_unix returns
+    # PAM_AUTHINFO_UNAVAIL before prompting, which means PAM_AUTHTOK is never
+    # set and the subsequent pam_sss use_first_pass call has nothing to work
+    # with — blocking LightDM and su logins even when IPA/SSSD auth succeeds.
+    hashedPassword = "!";
   };
 
-  # home-manager-<user>.service fails on first enrollment because /home/wayne
-  # doesn't exist until the user's first login (pam_mkhomedir creates it then).
-  # ConditionPathExists makes systemd skip the service (exit 0, condition not
-  # met) instead of failing. After first login the dir exists and subsequent
-  # rebuilds activate HM normally.
-  systemd.services."home-manager-${vars.ipaUser}".unitConfig.ConditionPathExists =
-    "/home/${vars.ipaUser}";
-
-  security.sudo.extraRules = [{
-    users = [ vars.ipaUser ];
-    commands = [{ command = "ALL"; options = [ "NOPASSWD" ]; }];
-  }];
-
   # Home Manager config for the IPA primary user, applied on every enrolled
   # host. Manages what IPA doesn't: dotfiles, user-scoped packages, session
   # variables. Switch-nix/Test-nix/buildImage are system-wide (configuration.nix)

modules/platforms/lxc.nix

@@ -52,6 +52,7 @@ in
   # LXC container does).
   imports = [
     (modulesPath + "/virtualisation/proxmox-lxc.nix")
+    ../common/preserve-ssh-host-key.nix
   ];
 
   proxmoxLXC = {
@@ -105,49 +106,6 @@ in
     };
   };
 
-  # NixOS's etc activation removes any /etc file that was in the previous
-  # generation's environment.etc but is absent from the current one — even
-  # real (non-symlink) copies.  On every routine nixos-rebuild switch/test that
-  # lacks NIXOS_HOST_KEYS_DIR the key is absent from environment.etc, so it
-  # gets removed as "obsolete".  sops-nix derives its age decryption key from
-  # /etc/ssh/ssh_host_ed25519_key; deletion cascades into every sops secret
-  # failing with "Error getting data key: 0 successful groups required, got 0".
-  #
-  # Fix: activation scripts that bracket the etc step, with explicit deps
-  # to enforce the correct ordering. Without deps the topological sort places
-  # preserveSshHostKey AFTER etc (confirmed live on a deployed lxc-tor-relay:
-  # position 7 vs etc's position 5) -- the key is already gone by the time it
-  # tries to save it. The etc/setupSecrets entries ADD to existing deps
-  # (types.listOf concatenates across module definitions).
-  system.activationScripts = {
-    # Saves the live key to /run before etc can delete it.
-    preserveSshHostKey = ''
-      if [ -f /etc/ssh/ssh_host_ed25519_key ]; then
-        cp /etc/ssh/ssh_host_ed25519_key /run/sshd-host-key-preserve.tmp
-        cp /etc/ssh/ssh_host_ed25519_key.pub /run/sshd-host-key-preserve.pub.tmp
-      fi
-    '';
-
-    # Reinstalls the key after etc runs if it was removed as "obsolete".
-    # The resulting file is not registered in environment.etc for either
-    # generation, so subsequent rebuilds leave it alone permanently.
-    restoreSshHostKey = {
-      deps = [ "etc" ];
-      text = ''
-        if [ ! -f /etc/ssh/ssh_host_ed25519_key ] && [ -f /run/sshd-host-key-preserve.tmp ]; then
-          install -m 0600 /run/sshd-host-key-preserve.tmp /etc/ssh/ssh_host_ed25519_key
-          install -m 0644 /run/sshd-host-key-preserve.pub.tmp /etc/ssh/ssh_host_ed25519_key.pub
-        fi
-        rm -f /run/sshd-host-key-preserve.tmp /run/sshd-host-key-preserve.pub.tmp
-      '';
-    };
-
-    # Force etc to wait until the key is saved, and sops to wait until the
-    # key is restored. Without these the topological sort breaks the chain.
-    etc = { deps = [ "preserveSshHostKey" ]; };
-    setupSecrets = { deps = [ "restoreSshHostKey" ]; };
-  };
-
   # virtualisation/proxmox-lxc.nix (imported above) registers the Nix
   # store DB via a systemd service (register-nix-paths) -- it never runs
   # an activation script at all. Confirmed live this means neither

modules/platforms/proxmox.nix

@@ -34,6 +34,7 @@ in
     ../hardware-configuration/vm/proxmox.nix
     ../boot/efi.nix
     ../disko/proxmox.nix
+    ../common/preserve-ssh-host-key.nix
   ];
 
   environment.etc = lib.mkIf hasKeyForThisTarget {
@@ -46,36 +47,4 @@ in
       mode = "0644";
     };
   };
-
-  # NixOS's etc activation removes any /etc file that was in the previous
-  # generation's environment.etc but is absent from the current one. Since
-  # the SSH key is only in environment.etc during the --impure build (when
-  # NIXOS_HOST_KEYS_DIR is set), normal rebuilds would remove it as
-  # "obsolete". These scripts mirror lxc.nix's approach: save the live key
-  # before etc runs, restore it after. Without the explicit deps, the
-  # topological sort places preserveSshHostKey after etc (confirmed live on
-  # lxc-tor-relay: position 7 vs etc's position 5), so the key is gone
-  # before it can be saved.
-  system.activationScripts = {
-    preserveSshHostKey = ''
-      if [ -f /etc/ssh/ssh_host_ed25519_key ]; then
-        cp /etc/ssh/ssh_host_ed25519_key /run/sshd-host-key-preserve.tmp
-        cp /etc/ssh/ssh_host_ed25519_key.pub /run/sshd-host-key-preserve.pub.tmp
-      fi
-    '';
-
-    restoreSshHostKey = {
-      deps = [ "etc" ];
-      text = ''
-        if [ ! -f /etc/ssh/ssh_host_ed25519_key ] && [ -f /run/sshd-host-key-preserve.tmp ]; then
-          install -m 0600 /run/sshd-host-key-preserve.tmp /etc/ssh/ssh_host_ed25519_key
-          install -m 0644 /run/sshd-host-key-preserve.pub.tmp /etc/ssh/ssh_host_ed25519_key.pub
-        fi
-        rm -f /run/sshd-host-key-preserve.tmp /run/sshd-host-key-preserve.pub.tmp
-      '';
-    };
-
-    etc = { deps = [ "preserveSshHostKey" ]; };
-    setupSecrets = { deps = [ "restoreSshHostKey" ]; };
-  };
 }

variables.nix

@@ -16,7 +16,7 @@
   serverIp = "192.168.2.226"; # server (NFS/ZFS) Proxmox VM LAN IP
   dockerIp = "192.168.2.225"; # docker Proxmox VM LAN IP
   pbsIp = "192.168.2.244"; # Proxmox Backup Server LAN IP (not NixOS-managed)
-  domainControllerIp = "192.168.2.253"; # FreeIPA domain controller / primary DNS (not NixOS-managed)
+  domainControllerIp = "192.168.2.253"; # FreeIPA domain controller — authoritative DNS for sweet.home (not NixOS-managed)
   ipaServer = "domain-controller.sweet.home"; # FreeIPA server hostname (used by security.ipa and Kerberos; must be a resolvable FQDN, not an IP)
 
   # Cross-host references (LAN hostnames/users other hosts reach over the network)
@@ -85,6 +85,12 @@
   # that IPA itself doesn't cover: dotfiles, user packages, session variables.
   ipaUser = "wayne";
 
+  # GID of the IPA "docker-access" group (GID 50010 on the IPA server).
+  # The local "docker" group is pinned to this GID on every host that runs
+  # Docker so that IPA group membership alone grants docker socket access -
+  # no per-host users.groups.docker.members entry for the IPA user needed.
+  dockerAccessGid = 50010;
+
   # HA file server cluster
   # LAN IPs (vmbr0 / ens18) — client-facing: iSCSI initiators, NFS, management.
   # Storage IPs (vmbr1 / ens19) — isolated internal bridge, used for DRBD