nixos/modules/build-types/server.nix

{ vars, lib, pkgs, ... }:

let
  poolName = lib.removePrefix "/" vars.storageRoot;

  # For each NFS share subpath, generate every ancestor path so ZFS datasets
  # are created parent-first.  e.g. "docker/config" → ["docker" "docker/config"]
  ancestors = path:
    let parts = lib.splitString "/" path;
    in lib.imap1 (i: _: lib.concatStringsSep "/" (lib.take i parts)) parts;

  poolDatasets = lib.unique (
    lib.concatMap (share: ancestors share.subpath)
      (lib.filter builtins.isAttrs (lib.attrValues vars.nfsShares))
  );
in
{
  imports = [
    ../beszel/enable-agent.nix
    ../services/zfs/enable-service.nix
  ];

  boot.zfs.extraPools = [ poolName ];

  # On a fresh image deploy the data disk (scsi1) starts blank — no pool
  # exists yet, so zfs-import-tank.service would spin for 60 s and fail.
  # This service runs first: if the pool is already present it exits instantly;
  # otherwise it creates it (with all required datasets) so the standard
  # import service finds it ready on the very first boot.
  systemd.services."zfs-init-${poolName}" = {
    description = "Initialize '${poolName}' ZFS pool on first boot if not present";
    wantedBy = [ "zfs-import-${poolName}.service" ];
    before = [ "zfs-import-${poolName}.service" ];
    after = [ "systemd-udev-settle.service" ];
    unitConfig.DefaultDependencies = false;
    serviceConfig = {
      Type = "oneshot";
      RemainAfterExit = true;
    };
    path = [ pkgs.zfs_unstable ];
    script = ''
      # Already imported — nothing to do.
      if zpool list "${poolName}" >/dev/null 2>&1; then
        exit 0
      fi

      # Locate the data disk first — used for both the fallback import
      # attempt and, only if the disk is genuinely blank, pool creation.
      DATA_DISK=""
      for candidate in /dev/disk/by-id/scsi-*drive-scsi1; do
        [[ "$candidate" == *-part* ]] && continue
        [ -b "$candidate" ] && DATA_DISK="$candidate" && break
      done

      if [ -z "$DATA_DISK" ]; then
        echo "zfs-init-${poolName}: no data disk found (expected /dev/disk/by-id/scsi-*drive-scsi1)" >&2
        exit 1
      fi

      # Try importing via the by-id symlink directory first (normal path),
      # then fall back to scanning the disk directly.  The two-step exists
      # because of a udev race: systemd-udev-settle.service can clear before
      # /dev/disk/by-id/ entries are fully populated, causing the first
      # import to fail even when the pool is intact on the disk.
      if zpool import -d /dev/disk/by-id -N "${poolName}" 2>/dev/null; then
        exit 0
      fi
      if zpool import -d "$DATA_DISK" -N "${poolName}" 2>/dev/null; then
        exit 0
      fi

      # Both import attempts failed.  Before creating a new pool, verify the
      # disk is genuinely blank — if ZFS label metadata is present the import
      # failed for some other reason and we must not clobber existing data.
      if zdb -l "$DATA_DISK" 2>/dev/null | grep -q "name: '${poolName}'"; then
        echo "zfs-init-${poolName}: $DATA_DISK has ZFS pool '${poolName}' metadata but import failed — refusing to overwrite existing data. Run 'zpool import -d $DATA_DISK ${poolName}' manually to investigate." >&2
        exit 1
      fi

      # Disk is genuinely blank: create the pool.  -f is intentionally
      # omitted so that if we somehow reach this point with an existing pool
      # on the disk, zpool refuses rather than silently destroying data.
      echo "zfs-init-${poolName}: creating pool on $DATA_DISK"
      zpool create "${poolName}" "$DATA_DISK"
      ${lib.concatMapStrings (ds: ''
        zfs create "${poolName}/${ds}"
      '') poolDatasets}
    '';
  };

  systemd.services.nfs-server = {
    after = [ "zfs-mount.service" ];
    requires = [ "zfs-mount.service" ];
  };

  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}
    '';
  };

  # mountd (20048) is needed for showmount/NFSv3 mount protocol — without it
  # clients can reach portmapper (111) and get the mountd port back, then
  # time out trying to connect to it. All three ports need TCP and UDP.
  networking.firewall.allowedTCPPorts = [ vars.ports.nfsRpcbind vars.ports.nfsd vars.ports.nfsMountd ];
  networking.firewall.allowedUDPPorts = [ vars.ports.nfsRpcbind vars.ports.nfsd vars.ports.nfsMountd ];
}