reimplement disko using the nixos type system

This should make the code cleaner, more robust and errors should be
clearer. we also changed the configuration format a bit.

1 files changed, 1058 insertions, 0 deletions

diff --git a/types.nix b/types.nix
new file mode 100644
index 0000000..e7dcf60
--- /dev/null
+++ b/types.nix
@@ -0,0 +1,1058 @@
+{ pkgs ? import <nixpkgs> {} }:
+with pkgs.lib;
+with builtins;
+
+rec {
+
+  diskoLib = {
+    # like types.oneOf but instead of a list takes an attrset
+    # uses the field "type" to find the correct type in the attrset
+    subType = typeAttr: mkOptionType rec {
+      name = "subType";
+      description = "one of ${attrNames typeAttr}";
+      check = x: typeAttr.${x.type}.check x;
+      merge = loc: defs:
+        foldl' (res: def: typeAttr.${def.value.type}.merge loc [def]) {} defs;
+      nestedTypes = typeAttr;
+    };
+
+    # option for valid contents of partitons (basically like devices, but without tables)
+    partitionType = mkOption {
+      type = types.nullOr (diskoLib.subType { inherit btrfs filesystem zfs mdraid luks lvm_pv; });
+      default = null;
+    };
+
+    # option for valid contents of devices
+    deviceType = mkOption {
+      type = types.nullOr (diskoLib.subType { inherit table btrfs filesystem zfs mdraid luks lvm_pv; });
+      default = null;
+    };
+
+    /* deepMergeMap takes a function and a list of attrsets and deep merges them
+
+       deepMergeMap :: -> (AttrSet -> AttrSet ) -> [ AttrSet ] -> Attrset
+
+       Example:
+         deepMergeMap (x: x.t = "test") [ { x = { y = 1; z = 3; }; } { x = { 123 = 234; }; } ]
+         => { x = { y = 1; z = 3; 123 = 234; t = "test"; }; }
+    */
+    deepMergeMap = f: listOfAttrs:
+      foldr (attr: acc: (recursiveUpdate acc (f attr))) {} listOfAttrs;
+
+    /* get a device and an index to get the matching device name
+
+       deviceNumbering :: str -> int -> str
+
+       Example:
+       deviceNumbering "/dev/sda" 3
+       => "/dev/sda3"
+
+       deviceNumbering "/dev/disk/by-id/xxx" 2
+       => "/dev/disk/by-id/xxx-part2"
+    */
+    deviceNumbering = dev: index:
+      let
+        schemas = {
+          dev__da = dev + toString index; # /dev/{s,v}da style
+          dev_disk = "${dev}-part${toString index}"; # /dev/disk/by-id/xxx style
+          dev_nvme = "${dev}n1p${toString index}"; # /dev/nvme0n1p1 style
+          dev_md = "${dev}p${toString index}"; # /dev/nvme0n1p1 style
+        };
+        detectSchema =
+          if match "/dev/[vs]d.*" dev != null then "dev__da" else
+          if match "/dev/disk/.*" dev != null then "dev_disk" else
+          if match "/dev/nvme.*" dev != null then "dev_nvme" else
+          if match "/dev/md/.*" dev != null then "dev_md" else
+          abort "${dev} seems not to be a supported disk format";
+      in schemas.${detectSchema};
+
+    /* Given a attrset of dependencies and a devices attrset
+       returns a sorted list by dependencies. aborts if a loop is found
+
+       sortDevicesByDependencies :: AttrSet -> AttrSet -> [ [ str str ] ]
+    */
+    sortDevicesByDependencies = dependencies: devices:
+      let
+        dependsOn = a: b:
+          elem a (attrByPath b [] dependencies);
+        maybeSortedDevices = toposort dependsOn (diskoLib.deviceList devices);
+      in
+        if (hasAttr "cycle" maybeSortedDevices) then
+          abort "detected a cycle in your disk setup: ${maybeSortedDevices.cycle}"
+        else
+          maybeSortedDevices.result;
+
+    /* Takes a devices attrSet and returns it as a list
+
+       deviceList :: AttrSet -> [ [ str str ] ]
+
+       Example:
+         deviceList { zfs.pool1 = {}; zfs.pool2 = {}; mdadm.raid1 = {}; }
+         => [ [ "zfs" "pool1" ] [ "zfs" "pool2" ] [ "mdadm" "raid1" ] ]
+    */
+    deviceList = devices:
+      concatLists (mapAttrsToList (n: v: (map (x: [ n x ]) (attrNames v))) devices);
+
+    /* Takes either a string or null and returns the string or an empty string
+
+       maybeStr :: Either (str null) -> str
+
+       Example:
+         maybeStr null
+         => ""
+         maybeSTr "hello world"
+         => "hello world"
+    */
+    maybeStr = x: optionalString (!isNull x) x;
+  };
+
+  optionTypes = rec {
+    # POSIX.1‐2017, 3.281 Portable Filename
+    filename = mkOptionType {
+      name = "POSIX portable filename";
+      check = x: isString x && builtins.match "[0-9A-Za-z._][0-9A-Za-z._-]*" x != null;
+      merge = mergeOneOption;
+    };
+
+    # POSIX.1‐2017, 3.2 Absolute Pathname
+    absolute-pathname = mkOptionType {
+      name = "POSIX absolute pathname";
+      check = x: isString x && substring 0 1 x == "/" && pathname.check x;
+      merge = mergeOneOption;
+    };
+
+    # POSIX.1-2017, 3.271 Pathname
+    pathname = mkOptionType {
+      name = "POSIX pathname";
+      check = x:
+        let
+          # The filter is used to normalize paths, i.e. to remove duplicated and
+          # trailing slashes.  It also removes leading slashes, thus we have to
+          # check for "/" explicitly below.
+          xs = filter (s: stringLength s > 0) (splitString "/" x);
+        in
+          isString x && (x == "/" || (length xs > 0 && all filename.check xs));
+      merge = mergeOneOption;
+    };
+  };
+
+  /* topLevel type of the disko config, takes attrsets of disks mdadms zpools and lvm vgs.
+     exports create, mount, meta and config
+  */
+  topLevel = types.submodule ({ config, ... }: {
+    options = {
+      devices = {
+        disk = mkOption {
+          type = types.attrsOf disk;
+          default = {};
+        };
+        mdadm = mkOption {
+          type = types.attrsOf mdadm;
+          default = {};
+        };
+        zpool = mkOption {
+          type = types.attrsOf zpool;
+          default = {};
+        };
+        lvm_vg = mkOption {
+          type = types.attrsOf lvm_vg;
+          default = {};
+        };
+      };
+      meta = mkOption {
+        readOnly = true;
+        default = diskoLib.deepMergeMap (dev: dev._meta) (flatten (map attrValues [
+          config.devices.disk
+          config.devices.lvm_vg
+          config.devices.mdadm
+          config.devices.zpool
+        ])) // {
+          sortedDeviceList = diskoLib.sortDevicesByDependencies config.meta.dependencies config.devices;
+        };
+      };
+      create = mkOption {
+        readOnly = true;
+        type = types.str;
+        default = ''
+          set -efux
+          ${concatStrings (map (dev: attrByPath (dev ++ [ "_create" ]) "" config.devices) config.meta.sortedDeviceList)}
+        '';
+      };
+      mount = mkOption {
+        readOnly = true;
+        type = types.str;
+        default = let
+          fsMounts = diskoLib.deepMergeMap (dev: dev._mount.fs or {}) (flatten (map attrValues [
+            config.devices.disk
+            config.devices.lvm_vg
+            config.devices.mdadm
+            config.devices.zpool
+          ]));
+        in ''
+          set -efux
+          # first create the neccessary devices
+          ${concatStrings (map (dev: attrByPath (dev ++ [ "_mount" "dev" ]) "" config.devices) config.meta.sortedDeviceList)}
+
+          # and then mount the filesystems in alphabetical order
+          # attrValues returns values sorted by name.  This is important, because it
+          # ensures that "/" is processed before "/foo" etc.
+          ${concatStrings (attrValues fsMounts)}
+        '';
+      };
+      config = mkOption {
+        readOnly = true;
+        default = diskoLib.deepMergeMap (dev: dev._config) (flatten (map attrValues [
+          config.devices.disk
+          config.devices.lvm_vg
+          config.devices.mdadm
+          config.devices.zpool
+        ]));
+      };
+    };
+  });
+
+  btrfs = types.submodule ({ config, ... }: {
+    options = {
+      type = mkOption {
+        type = types.enum [ "btrfs" ];
+        internal = true;
+      };
+      mountOptions = mkOption {
+        type = types.listOf types.str;
+        default = [];
+      };
+      subvolumes = mkOption {
+        type = types.listOf optionTypes.pathname;
+        default = [];
+      };
+      mountpoint = mkOption {
+        type = optionTypes.absolute-pathname;
+      };
+      _meta = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev: {
+        };
+      };
+      _create = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo types.str;
+        default = dev: ''
+          mkfs.btrfs ${dev}
+          ${optionalString (!isNull config.subvolumes or null) ''
+            MNTPOINT=$(mktemp -d)
+            (
+              mount ${dev} "$MNTPOINT"
+              trap 'umount $MNTPOINT; rm -rf $MNTPOINT' EXIT
+              ${concatMapStringsSep "\n" (subvolume: "btrfs subvolume create \"$MNTPOINT\"/${subvolume}") config.subvolumes}
+            )
+          ''}
+        '';
+      };
+      _mount = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev: {
+          fs.${config.mountpoint} = ''
+            if ! findmnt ${dev} "/mnt${config.mountpoint}" > /dev/null 2>&1; then
+              mount ${dev} "/mnt${config.mountpoint}" \
+              ${concatStringsSep " " config.mountOptions} \
+              -o X-mount.mkdir
+            fi
+          '';
+        };
+      };
+      _config = mkOption {
+        internal = true;
+        readOnly = true;
+        default = dev: {
+          fileSystems.${config.mountpoint} = {
+            device = dev;
+            fsType = "btrfs";
+          };
+        };
+      };
+    };
+  });
+
+  filesystem = types.submodule ({ config, ... }: {
+    options = {
+      type = mkOption {
+        type = types.enum [ "filesystem" ];
+        internal = true;
+      };
+      extraArgs = mkOption {
+        type = types.str;
+        default = "";
+      };
+      mountOptions = mkOption {
+        type = types.listOf types.str;
+        default = [];
+      };
+      options = mkOption {
+        type = types.listOf types.str;
+        default = [];
+      };
+      mountpoint = mkOption {
+        type = optionTypes.absolute-pathname;
+      };
+      format = mkOption {
+        type = types.str;
+      };
+      _meta = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev: {
+        };
+      };
+      _create = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo types.str;
+        default = dev: ''
+          mkfs.${config.format} \
+            ${config.extraArgs} \
+            ${dev}
+        '';
+      };
+      _mount = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev: {
+          fs.${config.mountpoint} = ''
+            if ! findmnt ${dev} "/mnt${config.mountpoint}" > /dev/null 2>&1; then
+              mount ${dev} "/mnt${config.mountpoint}" \
+              ${toString config.mountOptions} \
+              -o X-mount.mkdir
+            fi
+          '';
+        };
+      };
+      _config = mkOption {
+        internal = true;
+        readOnly = true;
+        default = dev: {
+          fileSystems.${config.mountpoint} = {
+            device = dev;
+            fsType = config.format;
+          };
+        };
+      };
+    };
+  });
+
+  table = types.submodule ({ config, ... }: {
+    options = {
+      type = mkOption {
+        type = types.enum [ "table" ];
+        internal = true;
+      };
+      format = mkOption {
+        type = types.enum [ "gpt" "msdos" ];
+        default = "gpt";
+      };
+      partitions = mkOption {
+        type = types.listOf partition;
+        default = [];
+      };
+      _meta = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev:
+          diskoLib.deepMergeMap (partition: partition._meta dev) config.partitions;
+      };
+      _create = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo types.str;
+        default = dev: ''
+          parted -s ${dev} mklabel ${config.format}
+          ${concatMapStrings (partition: partition._create dev config.format) config.partitions}
+        '';
+      };
+      _mount = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev:
+          let
+            partMounts = diskoLib.deepMergeMap (partition: partition._mount dev) config.partitions;
+          in {
+            dev = ''
+              ${concatStrings (map (x: x.dev or "") (attrValues partMounts))}
+            '';
+            fs = partMounts.fs or {};
+        };
+      };
+      _config = mkOption {
+        internal = true;
+        readOnly = true;
+        default = dev:
+          diskoLib.deepMergeMap (partition: partition._config dev) config.partitions;
+      };
+    };
+  });
+
+  partition = types.submodule ({ config, ... }: {
+    options = {
+      type = mkOption {
+        type = types.enum [ "partition" ];
+        internal = true;
+      };
+      part-type = mkOption {
+        type = types.enum [ "primary" "logical" "extended" ];
+        default = "primary";
+      };
+      fs-type = mkOption {
+        type = types.nullOr (types.enum [ "btrfs" "ext2" "ext3" "ext4" "fat16" "fat32" "hfs" "hfs+" "linux-swap" "ntfs" "reiserfs" "udf" "xfs" ]);
+        default = null;
+      };
+      name = mkOption {
+        type = types.nullOr types.str;
+        default = null;
+      };
+      start = mkOption {
+        type = types.str;
+        default = "0%";
+      };
+      end = mkOption {
+        type = types.str;
+        default = "100%";
+      };
+      index = mkOption {
+        type = types.int;
+        # TODO find a better way to get the index
+        default = toInt (head (match ".*entry ([[:digit:]]+)]" config._module.args.name));
+      };
+      flags = mkOption {
+        type = types.listOf types.str;
+        default = [];
+      };
+      bootable = mkOption {
+        type = types.bool;
+        default = false;
+      };
+      content = diskoLib.partitionType;
+      _meta = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev:
+          optionalAttrs (!isNull config.content) (config.content._meta dev);
+      };
+      _create = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (types.functionTo types.str);
+        default = dev: type: ''
+          ${optionalString (type == "gpt") ''
+            parted -s ${dev} mkpart ${config.name} ${diskoLib.maybeStr config.fs-type} ${config.start} ${config.end}
+          ''}
+          ${optionalString (type == "msdos") ''
+            parted -s ${dev} mkpart ${config.part-type} ${diskoLib.maybeStr config.fs-type} ${diskoLib.maybeStr config.fs-type} ${config.start} ${config.end}
+          ''}
+          # ensure /dev/disk/by-path/..-partN exists before continuing
+          udevadm trigger --subsystem-match=block; udevadm settle
+          ${optionalString (config.bootable) ''
+            parted -s ${dev} set ${toString config.index} boot on
+          ''}
+          ${concatMapStringsSep "" (flag: ''
+            parted -s ${dev} set ${toString config.index} ${flag} on
+          '') config.flags}
+          ${optionalString (!isNull config.content) (config.content._create (diskoLib.deviceNumbering dev config.index))}
+        '';
+      };
+      _mount = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev:
+          optionalAttrs (!isNull config.content) (config.content._mount (diskoLib.deviceNumbering dev config.index));
+      };
+      _config = mkOption {
+        internal = true;
+        readOnly = true;
+        default = dev:
+          optionalAttrs (!isNull config.content) (config.content._config (diskoLib.deviceNumbering dev config.index));
+      };
+    };
+  });
+
+  lvm_pv = types.submodule ({ config, ... }: {
+    options = {
+      type = mkOption {
+        type = types.enum [ "lvm_pv" ];
+        internal = true;
+      };
+      vg = mkOption {
+        type = types.str;
+      };
+      _meta = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev: {
+          dependencies.lvm_vg.${config.vg} = [ dev ];
+        };
+      };
+      _create = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo types.str;
+        default = dev: ''
+          pvcreate ${dev}
+          LVMDEVICES_${config.vg}="''${LVMDEVICES_${config.vg}:-}${dev} "
+        '';
+      };
+      _mount = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev:
+          {};
+      };
+      _config = mkOption {
+        internal = true;
+        readOnly = true;
+        default = dev: {};
+      };
+    };
+  });
+
+  lvm_vg = types.submodule ({ config, ... }: {
+    options = {
+      name = mkOption {
+        type = types.str;
+        default = config._module.args.name;
+      };
+      type = mkOption {
+        type = types.enum [ "lvm_vg" ];
+        internal = true;
+      };
+      lvs = mkOption {
+        type = types.attrsOf lvm_lv;
+        default = {};
+      };
+      _meta = mkOption {
+        internal = true;
+        readOnly = true;
+        type = (pkgs.formats.json {}).type;
+        default =
+          diskoLib.deepMergeMap (lv: lv._meta [ "lvm_vg" config.name ]) (attrValues config.lvs);
+      };
+      _create = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.str;
+        default = ''
+          vgcreate ${config.name} $LVMDEVICES_${config.name}
+          ${concatMapStrings (lv: lv._create config.name) (attrValues config.lvs)}
+        '';
+      };
+      _mount = mkOption {
+        internal = true;
+        readOnly = true;
+        type = (pkgs.formats.json {}).type;
+        default = let
+          lvMounts = diskoLib.deepMergeMap (lv: lv._mount config.name) (attrValues config.lvs);
+        in {
+          dev = ''
+            vgchange -a y
+            ${concatStrings (map (x: x.dev or "") (attrValues lvMounts))}
+          '';
+          fs = lvMounts.fs;
+        };
+      };
+      _config = mkOption {
+        internal = true;
+        readOnly = true;
+        default =
+          diskoLib.deepMergeMap (lv: lv._config config.name) (attrValues config.lvs);
+      };
+    };
+  });
+
+  lvm_lv = types.submodule ({ config, ... }: {
+    options = {
+      name = mkOption {
+        type = types.str;
+        default = config._module.args.name;
+      };
+      type = mkOption {
+        type = types.enum [ "lvm_lv" ];
+        default = "lvm_lv";
+        internal = true;
+      };
+      size = mkOption {
+        type = types.str; # TODO lvm size type
+      };
+      lvm_type = mkOption {
+        type = types.nullOr (types.enum [ "mirror" "raid0" "raid1" ]); # TODO add all types
+        default = null; # maybe there is always a default type?
+      };
+      extraArgs = mkOption {
+        type = types.str;
+        default = "";
+      };
+      content = diskoLib.partitionType;
+      _meta = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev:
+          optionalAttrs (!isNull config.content) (config.content._meta dev);
+      };
+      _create = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo types.str;
+        default = vg: ''
+          lvcreate \
+            ${if hasInfix "%" config.size then "-l" else "-L"} ${config.size} \
+            -n ${config.name} \
+            ${optionalString (!isNull config.lvm_type) "--type=${config.lvm_type}"} \
+            ${config.extraArgs} \
+            ${vg}
+          ${optionalString (!isNull config.content) (config.content._create "/dev/${vg}/${config.name}")}
+        '';
+      };
+      _mount = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = vg:
+          optionalAttrs (!isNull config.content) (config.content._mount "/dev/${vg}/${config.name}");
+      };
+      _config = mkOption {
+        internal = true;
+        readOnly = true;
+        default = dev:
+          optionalAttrs (!isNull config.content) (config.content._config dev);
+      };
+    };
+  });
+
+  zfs = types.submodule ({ config, ... }: {
+    options = {
+      type = mkOption {
+        type = types.enum [ "zfs" ];
+        internal = true;
+      };
+      pool = mkOption {
+        type = types.str;
+      };
+      _meta = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev: {
+          dependencies.zpool.${config.pool} = [ dev ];
+        };
+      };
+      _create = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo types.str;
+        default = dev: ''
+          ZFSDEVICES_${config.pool}="''${ZFSDEVICES_${config.pool}:-}${dev} "
+        '';
+      };
+      _mount = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev:
+          {};
+      };
+      _config = mkOption {
+        internal = true;
+        readOnly = true;
+        default = dev: {};
+      };
+    };
+  });
+
+  zpool = types.submodule ({ config, ... }: {
+    options = {
+      name = mkOption {
+        type = types.str;
+        default = config._module.args.name;
+      };
+      type = mkOption {
+        type = types.enum [ "zpool" ];
+        internal = true;
+      };
+      mode = mkOption {
+        type = types.str; # TODO zfs modes
+        default = "";
+      };
+      options = mkOption {
+        type = types.attrsOf types.str;
+        default = {};
+      };
+      rootFsOptions = mkOption {
+        type = types.attrsOf types.str;
+        default = {};
+      };
+      mountpoint = mkOption {
+        type = types.nullOr optionTypes.absolute-pathname;
+        default = null;
+      };
+      mountOptions = mkOption {
+        type = types.listOf types.str;
+        default = [];
+      };
+      datasets = mkOption {
+        type = types.attrsOf zfs_dataset;
+      };
+      _meta = mkOption {
+        internal = true;
+        readOnly = true;
+        type = (pkgs.formats.json {}).type;
+        default =
+          diskoLib.deepMergeMap (dataset: dataset._meta [ "zpool" config.name ]) (attrValues config.datasets);
+      };
+      _create = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.str;
+        default = ''
+          zpool create ${config.name} \
+            ${config.mode} \
+            ${concatStringsSep " " (mapAttrsToList (n: v: "-o ${n}=${v}") config.options)} \
+            ${concatStringsSep " " (mapAttrsToList (n: v: "-O ${n}=${v}") config.rootFsOptions)} \
+            ''${ZFSDEVICES_${config.name}}
+          ${concatMapStrings (dataset: dataset._create config.name) (attrValues config.datasets)}
+        '';
+      };
+      _mount = mkOption {
+        internal = true;
+        readOnly = true;
+        type = (pkgs.formats.json {}).type;
+        default = let
+          datasetMounts = diskoLib.deepMergeMap (dataset: dataset._mount config.name) (attrValues config.datasets);
+        in {
+          dev = ''
+            zpool list '${config.name}' >/dev/null 2>/dev/null || zpool import '${config.name}'
+            ${concatStrings (map (x: x.dev or "") (attrValues datasetMounts))}
+          '';
+          fs = datasetMounts.fs // optionalAttrs (!isNull config.mountpoint) {
+            ${config.mountpoint} = ''
+              if ! findmnt ${config.name} "/mnt${config.mountpoint}" > /dev/null 2>&1; then
+                mount ${config.name} "/mnt${config.mountpoint}" \
+                ${optionalString ((config.options.mountpoint or "") != "legacy") "-o zfsutil"} \
+                ${toString config.mountOptions} \
+                -o X-mount.mkdir \
+                -t zfs
+              fi
+            '';
+          };
+        };
+      };
+      _config = mkOption {
+        internal = true;
+        readOnly = true;
+        default =
+          recursiveUpdate
+            (diskoLib.deepMergeMap (dataset: dataset._config config.name) (attrValues config.datasets))
+            (optionalAttrs (!isNull config.mountpoint) {
+              fileSystems.${config.mountpoint} = {
+                device = config.name;
+                fsType = [ "zfs" ];
+              };
+            });
+      };
+    };
+  });
+
+  zfs_dataset = types.submodule ({ config, ... }: {
+    options = {
+      name = mkOption {
+        type = types.str;
+        default = config._module.args.name;
+      };
+      type = mkOption {
+        type = types.enum [ "zfs_dataset" ];
+        default = "zfs_dataset";
+      };
+      zfs_type = mkOption {
+        type = types.enum [ "filesystem" "volume" ];
+      };
+      options = mkOption {
+        type = types.attrsOf types.str;
+        default = {};
+      };
+      mountOptions = mkOption {
+        type = types.listOf types.str;
+        default = [];
+      };
+
+      # filesystem options
+      mountpoint = mkOption {
+        type = types.nullOr optionTypes.absolute-pathname;
+        default = null;
+      };
+
+      # volume options
+      size = mkOption {
+        type = types.nullOr types.str; # TODO size
+        default = null;
+      };
+
+      content = diskoLib.partitionType;
+      _meta = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev:
+          optionalAttrs (!isNull config.content) (config.content._meta dev);
+      };
+      _create = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo types.str;
+        default = zpool: ''
+          zfs create ${zpool}/${config.name} \
+            ${concatStringsSep " " (mapAttrsToList (n: v: "-o ${n}=${v}") config.options)} \
+            ${optionalString (config.zfs_type == "volume") "-V ${config.size}"}
+          ${optionalString (config.zfs_type == "volume") ''
+            udevadm trigger --subsystem-match=block; udevadm settle
+            ${optionalString (!isNull config.content) (config.content._create "/dev/zvol/${zpool}/${config.name}")}
+          ''}
+        '';
+      };
+      _mount = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = zpool:
+          optionalAttrs (config.zfs_type == "volume" && !isNull config.content) (config.content._mount "/dev/zvol/${zpool}/${config.name}") //
+            optionalAttrs (config.zfs_type == "filesystem" && config.options.mountpoint or "" != "none") { fs.${config.mountpoint} = ''
+              if ! findmnt ${zpool}/${config.name} "/mnt${config.mountpoint}" > /dev/null 2>&1; then
+                mount ${zpool}/${config.name} "/mnt${config.mountpoint}" \
+                -o X-mount.mkdir \
+                ${toString config.mountOptions} \
+                ${optionalString ((config.options.mountpoint or "") != "legacy") "-o zfsutil"} \
+                -t zfs
+              fi
+            ''; };
+      };
+      _config = mkOption {
+        internal = true;
+        readOnly = true;
+        default = zpool:
+          optionalAttrs (config.zfs_type == "volume" && !isNull config.content) (config.content._config "/dev/zvol/${zpool}/${config.name}") //
+            optionalAttrs (config.zfs_type == "filesystem" && config.options.mountpoint or "" != "none") {
+              fileSystems.${config.mountpoint} = {
+                device = "${zpool}/${config.name}";
+                fsType = [ "zfs" ];
+              };
+            };
+      };
+    };
+  });
+
+  mdadm = types.submodule ({ config, ... }: {
+    options = {
+      name = mkOption {
+        type = types.str;
+        default = config._module.args.name;
+      };
+      type = mkOption {
+        type = types.enum [ "mdadm" ];
+        default = "mdadm";
+      };
+      level = mkOption {
+        type = types.int;
+        default = 1;
+      };
+      content = diskoLib.deviceType;
+      _meta = mkOption {
+        internal = true;
+        readOnly = true;
+        type = (pkgs.formats.json {}).type;
+        default =
+          optionalAttrs (!isNull config.content) (config.content._meta [ "mdadm" config.name ]);
+      };
+      _create = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.str;
+        default = ''
+          echo 'y' | mdadm --create /dev/md/${config.name} \
+            --level=${toString config.level} \
+            --raid-devices=''${RAIDDEVICES_N_${config.name}} \
+            ''${RAIDDEVICES_${config.name}}
+          udevadm trigger --subsystem-match=block; udevadm settle
+          ${optionalString (!isNull config.content) (config.content._create "/dev/md/${config.name}")}
+        '';
+      };
+      _mount = mkOption {
+        internal = true;
+        readOnly = true;
+        type = (pkgs.formats.json {}).type;
+        default =
+          optionalAttrs (!isNull config.content) (config.content._mount "/dev/md/${config.name}");
+        # TODO we probably need to assemble the mdadm somehow
+      };
+      _config = mkOption {
+        internal = true;
+        readOnly = true;
+        default =
+          optionalAttrs (!isNull config.content) (config.content._config "/dev/md/${config.name}");
+      };
+    };
+  });
+
+  mdraid = types.submodule ({ config, ... }: {
+    options = {
+      type = mkOption {
+        type = types.enum [ "mdraid" ];
+      };
+
+      name = mkOption {
+        type = types.str;
+      };
+      _meta = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev: {
+          dependencies.mdadm.${config.name} = [ dev ];
+        };
+      };
+      _create = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo types.str;
+        default = dev: ''
+          RAIDDEVICES_N_${config.name}=$((''${RAIDDEVICES_N_${config.name}:-0}+1))
+          RAIDDEVICES_${config.name}="''${RAIDDEVICES_${config.name}:-}${dev} "
+        '';
+      };
+      _mount = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev:
+          {};
+      };
+      _config = mkOption {
+        internal = true;
+        readOnly = true;
+        default = dev: {};
+      };
+    };
+  });
+
+  luks = types.submodule ({ config, ... }: {
+    options = {
+      type = mkOption {
+        type = types.enum [ "luks" ];
+      };
+      name = mkOption {
+        type = types.str;
+      };
+      keyFile = mkOption {
+        type = types.nullOr optionTypes.absolute-pathname;
+        default = null;
+      };
+      extraArgs = mkOption {
+        type = types.listOf types.str;
+        default = [];
+      };
+      content = diskoLib.deviceType;
+      _meta = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev:
+          optionalAttrs (!isNull config.content) (config.content._meta dev);
+      };
+      _create = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo types.str;
+        default = dev: ''
+          cryptsetup -q luksFormat ${dev} ${diskoLib.maybeStr config.keyFile} ${toString config.extraArgs}
+          cryptsetup luksOpen ${dev} ${config.name} ${optionalString (!isNull config.keyFile) "--key-file ${config.keyFile}"}
+          ${optionalString (!isNull config.content) (config.content._create "/dev/mapper/${config.name}")}
+        '';
+      };
+      _mount = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.functionTo (pkgs.formats.json {}).type;
+        default = dev:
+          let
+            contentMount = config.content._mount "/dev/mapper/${config.name}";
+          in
+            {
+              dev = ''
+                cryptsetup status ${config.name} >/dev/null 2>/dev/null ||
+                  cryptsetup luksOpen ${dev} ${config.name} ${optionalString (!isNull config.keyFile) "--key-file ${config.keyFile}"}
+                ${optionalString (!isNull config.content) contentMount.dev or ""}
+              '';
+              fs = optionalAttrs (!isNull config.content) contentMount.fs or {};
+            };
+      };
+      _config = mkOption {
+        internal = true;
+        readOnly = true;
+        default = dev:
+          recursiveUpdate {
+            # TODO do we need this always in initrd and only there?
+            boot.initrd.luks.devices.${config.name}.device = dev;
+          } (optionalAttrs (!isNull config.content) (config.content._config "/dev/mapper/${config.name}"));
+      };
+    };
+  });
+
+  disk = types.submodule ({ config, ... }: {
+    options = {
+      name = mkOption {
+        type = types.str;
+        default = config._module.args.name;
+      };
+      type = mkOption {
+        type = types.enum [ "disk" ];
+      };
+      device = mkOption {
+        type = optionTypes.absolute-pathname; # TODO check if subpath of /dev ?
+      };
+      content = diskoLib.deviceType;
+      _meta = mkOption {
+        internal = true;
+        readOnly = true;
+        type = (pkgs.formats.json {}).type;
+        default =
+          optionalAttrs (!isNull config.content) (config.content._meta [ "disk" config.device ]);
+      };
+      _create = mkOption {
+        internal = true;
+        readOnly = true;
+        type = types.str;
+        default = config.content._create config.device;
+      };
+      _mount = mkOption {
+        internal = true;
+        readOnly = true;
+        type = (pkgs.formats.json {}).type;
+        default =
+          optionalAttrs (!isNull config.content) (config.content._mount config.device);
+      };
+      _config = mkOption {
+        internal = true;
+        readOnly = true;
+        default =
+          optionalAttrs (!isNull config.content) (config.content._config config.device);
+      };
+    };
+  });
+}