feat: add unified microvm & container definition; add net, misc, disko lib extension

2025-10-10 22:00:39 +02:00 · 2024-01-11 02:56:19 +01:00 · 2024-01-11 02:56:19 +01:00 · c55f465ba1
commit c55f465ba1
parent 6e0ea91254
13 changed files with 1093 additions and 3 deletions
--- a/README.md
+++ b/README.md
@ -35,3 +35,6 @@ Certain modules may require the use of additional flakes. In particular you migh
 - [impermanence](https://github.com/nix-community/impermanence)
 - [agenix](https://github.com/ryantm/agenix)
 - [agenix-rekey](https://github.com/oddlama/agenix-rekey)
+- [microvm.nix](https://github.com/astro/microvm.nix)
+
+You also must have a `specialArgs.inputs` that refers to all of your flake's inputs.
--- a/flake.lock
+++ b/flake.lock
@ -76,6 +76,19 @@
        "type": "github"
      }
    },
+    "lib-net": {
+      "flake": false,
+      "locked": {
+        "lastModified": 1596309860,
+        "narHash": "sha256-izAzepR/6cDvnRfaa2ceSolMLMwqzQB5x9q62aR5J2g=",
+        "type": "tarball",
+        "url": "https://gist.github.com/duairc/5c9bb3c922e5d501a1edb9e7b3b845ba/archive/3885f7cd9ed0a746a9d675da6f265d41e9fd6704.tar.gz"
+      },
+      "original": {
+        "type": "tarball",
+        "url": "https://gist.github.com/duairc/5c9bb3c922e5d501a1edb9e7b3b845ba/archive/3885f7cd9ed0a746a9d675da6f265d41e9fd6704.tar.gz"
+      }
+    },
    "nixpkgs": {
      "locked": {
        "lastModified": 1703013332,
@ -138,6 +151,7 @@
      "inputs": {
        "devshell": "devshell",
        "flake-utils": "flake-utils",
+        "lib-net": "lib-net",
        "nixpkgs": "nixpkgs",
        "pre-commit-hooks": "pre-commit-hooks"
      }
--- a/flake.nix
+++ b/flake.nix
@ -6,6 +6,12 @@
    };

    flake-utils.url = "github:numtide/flake-utils";
+
+    lib-net = {
+      url = "https://gist.github.com/duairc/5c9bb3c922e5d501a1edb9e7b3b845ba/archive/3885f7cd9ed0a746a9d675da6f265d41e9fd6704.tar.gz";
+      flake = false;
+    };
+
    nixpkgs.url = "github:NixOS/nixpkgs/nixos-unstable";

    pre-commit-hooks = {
@ -23,12 +29,14 @@
    devshell,
    pre-commit-hooks,
    ...
-  }:
+  } @ inputs:
    {
      nixosModules.extra-modules = import ./modules;
      nixosModules.default = self.nixosModules.extra-modules;
      homeManagerModules.extra-modules = import ./hm-modules;
      homeManagerModules.default = self.homeManagerModules.extra-modules;
+      overlays.extra-modules = import ./lib inputs;
+      overlays.default = self.overlays.extra-modules;
    }
    // flake-utils.lib.eachDefaultSystem (system: rec {
      pkgs = import nixpkgs {
--- a/lib/default.nix
+++ b/lib/default.nix
@ -0,0 +1,17 @@
+inputs: final: prev:
+prev.lib.composeManyExtensions (
+  # Order is important to allow using prev instead of final in more places to
+  # speed up evaluation.
+  map (x: import x inputs) [
+    # No dependencies
+    ./types.nix
+    # No dependencies
+    ./misc.nix
+    # No dependencies
+    ./disko.nix
+    # Requires misc
+    ./net.nix
+  ]
+)
+final
+prev
--- a/lib/disko.nix
+++ b/lib/disko.nix
@ -0,0 +1,89 @@
+_inputs: final: prev: {
+  lib =
+    prev.lib
+    // {
+      disko = {
+        content = {
+          luksZfs = luksName: pool: {
+            type = "luks";
+            name = "${pool}_${luksName}";
+            settings.allowDiscards = true;
+            content = {
+              type = "zfs";
+              inherit pool;
+            };
+          };
+        };
+        gpt = {
+          partGrub = name: start: end: {
+            inherit name start end;
+            part-type = "primary";
+            flags = ["bios_grub"];
+          };
+          partEfi = name: start: end: {
+            inherit name start end;
+            fs-type = "fat32";
+            bootable = true;
+            content = {
+              type = "filesystem";
+              format = "vfat";
+              mountpoint = "/boot";
+            };
+          };
+          partSwap = name: start: end: {
+            inherit name start end;
+            fs-type = "linux-swap";
+            content = {
+              type = "swap";
+              randomEncryption = true;
+            };
+          };
+          partLuksZfs = luksName: pool: start: end: {
+            inherit start end;
+            name = "${pool}_${luksName}";
+            content = final.lib.disko.content.luksZfs luksName pool;
+          };
+        };
+        zfs = rec {
+          mkZpool = prev.lib.recursiveUpdate {
+            type = "zpool";
+            rootFsOptions = {
+              compression = "zstd";
+              acltype = "posix";
+              atime = "off";
+              xattr = "sa";
+              dnodesize = "auto";
+              mountpoint = "none";
+              canmount = "off";
+              devices = "off";
+            };
+            options.ashift = "12";
+          };
+
+          impermanenceZfsDatasets = {
+            "local" = unmountable;
+            "local/root" =
+              filesystem "/"
+              // {
+                postCreateHook = "zfs snapshot rpool/local/root@blank";
+              };
+            "local/nix" = filesystem "/nix";
+            "local/state" = filesystem "/state";
+            "safe" = unmountable;
+            "safe/persist" = filesystem "/persist";
+          };
+
+          unmountable = {type = "zfs_fs";};
+          filesystem = mountpoint: {
+            type = "zfs_fs";
+            options = {
+              canmount = "noauto";
+              inherit mountpoint;
+            };
+            # Required to add dependencies for initrd
+            inherit mountpoint;
+          };
+        };
+      };
+    };
+}
--- a/lib/misc.nix
+++ b/lib/misc.nix
@ -0,0 +1,84 @@
+_inputs: _final: prev: let
+  inherit
+    (prev.lib)
+    filter
+    foldl'
+    genAttrs
+    genList
+    mergeAttrs
+    mkMerge
+    stringToCharacters
+    substring
+    unique
+    ;
+
+  # Counts how often each element occurrs in xs.
+  # Elements must be strings.
+  countOccurrences =
+    foldl'
+    (acc: x: acc // {${x} = (acc.${x} or 0) + 1;})
+    {};
+
+  # Returns all elements in xs that occur at least twice
+  duplicates = xs: let
+    occurrences = countOccurrences xs;
+  in
+    unique (filter (x: occurrences.${x} > 1) xs);
+
+  # Concatenates all given attrsets as if calling a // b in order.
+  concatAttrs = foldl' mergeAttrs {};
+
+  # True if the path or string starts with /
+  isAbsolutePath = x: substring 0 1 x == "/";
+
+  # Merges all given attributes from the given attrsets using mkMerge.
+  # Useful to merge several top-level configs in a module.
+  mergeToplevelConfigs = keys: attrs:
+    genAttrs keys (attr: mkMerge (map (x: x.${attr} or {}) attrs));
+
+  # Calculates base^exp, but careful, this overflows for results > 2^62
+  pow = base: exp: foldl' (a: x: x * a) 1 (genList (_: base) exp);
+
+  hexLiteralValues = {
+    "0" = 0;
+    "1" = 1;
+    "2" = 2;
+    "3" = 3;
+    "4" = 4;
+    "5" = 5;
+    "6" = 6;
+    "7" = 7;
+    "8" = 8;
+    "9" = 9;
+    "a" = 10;
+    "b" = 11;
+    "c" = 12;
+    "d" = 13;
+    "e" = 14;
+    "f" = 15;
+    "A" = 10;
+    "B" = 11;
+    "C" = 12;
+    "D" = 13;
+    "E" = 14;
+    "F" = 15;
+  };
+
+  # Converts the given hex string to an integer. Only reliable for inputs in [0, 2^63),
+  # after that the sign bit will overflow.
+  hexToDec = v: foldl' (acc: x: acc * 16 + hexLiteralValues.${x}) 0 (stringToCharacters v);
+in {
+  lib =
+    prev.lib
+    // {
+      inherit
+        concatAttrs
+        countOccurrences
+        duplicates
+        hexToDec
+        isAbsolutePath
+        mergeToplevelConfigs
+        pow
+        ;
+    };
+}
--- a/lib/net.nix
+++ b/lib/net.nix
@ -0,0 +1,361 @@
+inputs: _final: prev: let
+  inherit
+    (prev.lib)
+    all
+    any
+    assertMsg
+    elem
+    filter
+    flip
+    foldl'
+    hasInfix
+    head
+    min
+    partition
+    range
+    recursiveUpdate
+    reverseList
+    splitString
+    substring
+    unique
+    warnIf
+    ;
+
+  # From misc.nix
+  inherit
+    (prev.lib)
+    hexToDec
+    pow
+    ;
+
+  # IP address math library
+  # https://gist.github.com/duairc/5c9bb3c922e5d501a1edb9e7b3b845ba
+  # Plus some extensions by us
+  libNet =
+    (import "${inputs.lib-net}/net.nix" {
+      inherit (inputs.nixpkgs) lib;
+    })
+    .lib
+    .net;
+in {
+  lib = recursiveUpdate prev.lib {
+    net = recursiveUpdate (removeAttrs libNet ["types"]) {
+      cidr = rec {
+        # host :: (ip | mac | integer) -> cidr -> ip
+        #
+        # Wrapper that extends the original host function to
+        # check whether the argument `n` is in-range for the given cidr.
+        #
+        # Examples:
+        #
+        # > net.cidr.host 255 "192.168.1.0/24"
+        # "192.168.1.255"
+        # > net.cidr.host (256) "192.168.1.0/24"
+        # <fails with an error message>
+        # > net.cidr.host (-1) "192.168.1.0/24"
+        # "192.168.1.255"
+        # > net.cidr.host (-256) "192.168.1.0/24"
+        # "192.168.1.0"
+        # > net.cidr.host (-257) "192.168.1.0/24"
+        # <fails with an error message>
+        host = i: n: let
+          cap = libNet.cidr.capacity n;
+        in
+          assert assertMsg (i >= (-cap) && i < cap) "The host ${toString i} lies outside of ${n}";
+            libNet.cidr.host i n;
+        # hostCidr :: (ip | mac | integer) -> cidr -> cidr
+        #
+        # Returns the nth host in the given cidr range (like cidr.host)
+        # but as a cidr that retains the original prefix length.
+        #
+        # Examples:
+        #
+        # > net.cidr.hostCidr 2 "192.168.1.0/24"
+        # "192.168.1.2/24"
+        hostCidr = n: x: "${libNet.cidr.host n x}/${toString (libNet.cidr.length x)}";
+        # ip :: (cidr | ip) -> ip
+        #
+        # Returns just the ip part of the cidr.
+        #
+        # Examples:
+        #
+        # > net.cidr.ip "192.168.1.100/24"
+        # "192.168.1.100"
+        # > net.cidr.ip "192.168.1.100"
+        # "192.168.1.100"
+        ip = x: head (splitString "/" x);
+        # canonicalize :: cidr -> cidr
+        #
+        # Replaces the ip of the cidr with the canonical network address
+        # (first contained address in range)
+        #
+        # Examples:
+        #
+        # > net.cidr.canonicalize "192.168.1.100/24"
+        # "192.168.1.0/24"
+        canonicalize = x: libNet.cidr.make (libNet.cidr.length x) (ip x);
+        # mergev4 :: [cidrv4 | ipv4] -> (cidrv4 | null)
+        #
+        # Returns the smallest cidr network that includes all given networks.
+        # If no cidr mask is given, /32 is assumed.
+        #
+        # Examples:
+        #
+        # > net.cidr.mergev4 ["192.168.1.1/24" "192.168.6.1/32"]
+        # "192.168.0.0/21"
+        mergev4 = addrs_: let
+          # Append /32 if necessary
+          addrs = map (x:
+            if hasInfix "/" x
+            then x
+            else "${x}/32")
+          addrs_;
+          # The smallest occurring length is the first we need to start checking, since
+          # any greater cidr length represents a smaller address range which
+          # wouldn't contain all of the original addresses.
+          startLength = foldl' min 32 (map libNet.cidr.length addrs);
+          possibleLengths = reverseList (range 0 startLength);
+          # The first ip address will be "expanded" in cidr length until it covers all other
+          # used addresses.
+          firstIp = ip (head addrs);
+          # Return the first (i.e. greatest length -> smallest prefix) cidr length
+          # in the list that covers all used addresses
+          bestLength = head (filter
+            # All given addresses must be contained by the generated address.
+            (len:
+              all (x:
+                libNet.cidr.contains
+                (ip x)
+                (libNet.cidr.make len firstIp))
+              addrs)
+            possibleLengths);
+        in
+          assert assertMsg (!any (hasInfix ":") addrs) "mergev4 cannot operate on ipv6 addresses";
+            if addrs == []
+            then null
+            else libNet.cidr.make bestLength firstIp;
+        # mergev6 :: [cidrv6 | ipv6] -> (cidrv6 | null)
+        #
+        # Returns the smallest cidr network that includes all given networks.
+        # If no cidr mask is given, /128 is assumed.
+        #
+        # Examples:
+        #
+        # > net.cidr.mergev6 ["fd00:dead:cafe::/64" "fd00:fd12:3456:7890::/56"]
+        # "fd00:c000::/18"
+        mergev6 = addrs_: let
+          # Append /128 if necessary
+          addrs = map (x:
+            if hasInfix "/" x
+            then x
+            else "${x}/128")
+          addrs_;
+          # The smallest occurring length is the first we need to start checking, since
+          # any greater cidr length represents a smaller address range which
+          # wouldn't contain all of the original addresses.
+          startLength = foldl' min 128 (map libNet.cidr.length addrs);
+          possibleLengths = reverseList (range 0 startLength);
+          # The first ip address will be "expanded" in cidr length until it covers all other
+          # used addresses.
+          firstIp = ip (head addrs);
+          # Return the first (i.e. greatest length -> smallest prefix) cidr length
+          # in the list that covers all used addresses
+          bestLength = head (filter
+            # All given addresses must be contained by the generated address.
+            (len:
+              all (x:
+                libNet.cidr.contains
+                (ip x)
+                (libNet.cidr.make len firstIp))
+              addrs)
+            possibleLengths);
+        in
+          assert assertMsg (all (hasInfix ":") addrs) "mergev6 cannot operate on ipv4 addresses";
+            if addrs == []
+            then null
+            else libNet.cidr.make bestLength firstIp;
+        # merge :: [cidr] -> { cidrv4 = (cidrv4 | null); cidrv6 = (cidrv4 | null); }
+        #
+        # Returns the smallest cidr network that includes all given networks,
+        # but yields two separate result for all given ipv4 and ipv6 addresses.
+        # Equivalent to calling mergev4 and mergev6 on a partition individually.
+        merge = addrs: let
+          v4_and_v6 = partition (hasInfix ":") addrs;
+        in {
+          cidrv4 = mergev4 v4_and_v6.wrong;
+          cidrv6 = mergev6 v4_and_v6.right;
+        };
+        # assignIps :: cidr -> [int | ip] -> [string] -> [ip]
+        #
+        # Assigns a semi-stable ip address from the given cidr network to each hostname.
+        # The algorithm is based on hashing (abusing sha256) with linear probing.
+        # The order of hosts doesn't matter. No ip (or offset) from the reserved list
+        # will be assigned. The network address and broadcast address will always be reserved
+        # automatically.
+        #
+        # Examples:
+        #
+        # > net.cidr.assignIps "192.168.100.1/24" [] ["a" "b" "c"]
+        # { a = "192.168.100.202"; b = "192.168.100.74"; c = "192.168.100.226"; }
+        #
+        # > net.cidr.assignIps "192.168.100.1/24" [] ["a" "b" "c" "a-new-elem"]
+        # { a = "192.168.100.202"; a-new-elem = "192.168.100.88"; b = "192.168.100.74"; c = "192.168.100.226"; }
+        #
+        # > net.cidr.assignIps "192.168.100.1/24" [202 "192.168.100.74"] ["a" "b" "c"]
+        # { a = "192.168.100.203"; b = "192.168.100.75"; c = "192.168.100.226"; }
+        assignIps = net: reserved: hosts: let
+          cidrSize = libNet.cidr.size net;
+          capacity = libNet.cidr.capacity net;
+          # The base address of the network. Used to convert ip-based reservations to offsets
+          baseAddr = host 0 net;
+          # Reserve some values for the network, host and broadcast address.
+          # The network and broadcast address should never be used, and we
+          # want to reserve the host address for the host. We also convert
+          # any ips to offsets here.
+          init = unique (
+            [0 (capacity - 1)]
+            ++ flip map reserved (x:
+              if builtins.typeOf x == "int"
+              then x
+              else -(libNet.ip.diff baseAddr x))
+          );
+          nHosts = builtins.length hosts;
+          nInit = builtins.length init;
+          # Pre-sort all hosts, to ensure ordering invariance
+          sortedHosts =
+            warnIf
+            ((nInit + nHosts) > 0.3 * capacity)
+            "assignIps: hash stability may be degraded since utilization is >30%"
+            (builtins.sort builtins.lessThan hosts);
+          # Generates a hash (i.e. offset value) for a given hostname
+          hashElem = x:
+            builtins.bitAnd (capacity - 1)
+            (hexToDec (builtins.substring 0 16 (builtins.hashString "sha256" x)));
+          # Do linear probing. Returns the first unused value at or after the given value.
+          probe = avoid: value:
+            if elem value avoid
+            # TODO lib.mod
+            # Poor man's modulo, because nix has no modulo. Luckily we operate on a residue
+            # class of x modulo 2^n, so we can use bitAnd instead.
+            then probe avoid (builtins.bitAnd (capacity - 1) (value + 1))
+            else value;
+          # Hash a new element and avoid assigning any existing values.
+          assignOne = {
+            assigned,
+            used,
+          }: x: let
+            value = probe used (hashElem x);
+          in {
+            assigned =
+              assigned
+              // {
+                ${x} = host value net;
+              };
+            used = [value] ++ used;
+          };
+        in
+          assert assertMsg (cidrSize >= 2 && cidrSize <= 62)
+          "assignIps: cidrSize=${toString cidrSize} is not in [2, 62].";
+          assert assertMsg (nHosts <= capacity - nInit)
+          "assignIps: number of hosts (${toString nHosts}) must be <= capacity (${toString capacity}) - reserved (${toString nInit})";
+          # Assign an ip in the subnet to each element, in order
+            (foldl' assignOne {
+                assigned = {};
+                used = init;
+              }
+              sortedHosts)
+            .assigned;
+      };
+      ip = rec {
+        # Checks whether the given address (with or without cidr notation) is an ipv4 address.
+        isv4 = x: !isv6 x;
+        # Checks whether the given address (with or without cidr notation) is an ipv6 address.
+        isv6 = hasInfix ":";
+      };
+      mac = {
+        # Adds offset to the given base address and ensures the result is in
+        # a locally administered range by replacing the second nibble with a 2.
+        addPrivate = base: offset: let
+          added = libNet.mac.add base offset;
+          pre = substring 0 1 added;
+          suf = substring 2 (-1) added;
+        in "${pre}2${suf}";
+        # assignMacs :: mac (base) -> int (size) -> [int | mac] (reserved) -> [string] (hosts) -> [mac]
+        #
+        # Assigns a semi-stable MAC address starting in [base, base + 2^size) to each hostname.
+        # The algorithm is based on hashing (abusing sha256) with linear probing.
+        # The order of hosts doesn't matter. No mac (or offset) from the reserved list
+        # will be assigned.
+        #
+        # Examples:
+        #
+        # > net.mac.assignMacs "11:22:33:00:00:00" 24 [] ["a" "b" "c"]
+        # { a = "11:22:33:1b:bd:ca"; b = "11:22:33:39:59:4a"; c = "11:22:33:50:7a:e2"; }
+        #
+        # > net.mac.assignMacs "11:22:33:00:00:00" 24 [] ["a" "b" "c" "a-new-elem"]
+        # { a = "11:22:33:1b:bd:ca"; a-new-elem = "11:22:33:d6:5d:58"; b = "11:22:33:39:59:4a"; c = "11:22:33:50:7a:e2"; }
+        #
+        # > net.mac.assignMacs "11:22:33:00:00:00" 24 ["11:22:33:1b:bd:ca"] ["a" "b" "c"]
+        # { a = "11:22:33:1b:bd:cb"; b = "11:22:33:39:59:4a"; c = "11:22:33:50:7a:e2"; }
+        assignMacs = base: size: reserved: hosts: let
+          capacity = pow 2 size;
+          baseAsInt = libNet.mac.diff base "00:00:00:00:00:00";
+          init = unique (
+            flip map reserved (x:
+              if builtins.typeOf x == "int"
+              then x
+              else libNet.mac.diff x base)
+          );
+          nHosts = builtins.length hosts;
+          nInit = builtins.length init;
+          # Pre-sort all hosts, to ensure ordering invariance
+          sortedHosts =
+            warnIf
+            ((nInit + nHosts) > 0.3 * capacity)
+            "assignMacs: hash stability may be degraded since utilization is >30%"
+            (builtins.sort builtins.lessThan hosts);
+          # Generates a hash (i.e. offset value) for a given hostname
+          hashElem = x:
+            builtins.bitAnd (capacity - 1)
+            (hexToDec (substring 0 16 (builtins.hashString "sha256" x)));
+          # Do linear probing. Returns the first unused value at or after the given value.
+          probe = avoid: value:
+            if elem value avoid
+            # TODO lib.mod
+            # Poor man's modulo, because nix has no modulo. Luckily we operate on a residue
+            # class of x modulo 2^n, so we can use bitAnd instead.
+            then probe avoid (builtins.bitAnd (capacity - 1) (value + 1))
+            else value;
+          # Hash a new element and avoid assigning any existing values.
+          assignOne = {
+            assigned,
+            used,
+          }: x: let
+            value = probe used (hashElem x);
+          in {
+            assigned =
+              assigned
+              // {
+                ${x} = libNet.mac.add value base;
+              };
+            used = [value] ++ used;
+          };
+        in
+          assert assertMsg (size >= 2 && size <= 62)
+          "assignMacs: size=${toString size} is not in [2, 62].";
+          assert assertMsg (builtins.bitAnd (capacity - 1) baseAsInt == 0)
+          "assignMacs: the size=${toString size} least significant bits of the base mac address must be 0.";
+          assert assertMsg (nHosts <= capacity - nInit)
+          "assignMacs: number of hosts (${toString nHosts}) must be <= capacity (${toString capacity}) - reserved (${toString nInit})";
+          # Assign an ip in the subnet to each element, in order
+            (foldl' assignOne {
+                assigned = {};
+                used = init;
+              }
+              sortedHosts)
+            .assigned;
+      };
+    };
+    types.net = libNet.types;
+  };
+}
--- a/lib/types.nix
+++ b/lib/types.nix
@ -0,0 +1,52 @@
+_inputs: _final: prev: let
+  inherit
+    (prev.lib)
+    all
+    assertMsg
+    isAttrs
+    mkOptionType
+    recursiveUpdate
+    showOption
+    types
+    ;
+
+  # Checks whether the value is a lazy value without causing
+  # it's value to be evaluated
+  isLazyValue = x: isAttrs x && x ? _lazyValue;
+  # Constructs a lazy value holding the given value.
+  lazyValue = value: {_lazyValue = value;};
+
+  # Represents a lazy value of the given type, which
+  # holds the actual value as an attrset like { _lazyValue = <actual value>; }.
+  # This allows the option to be defined and filtered from a defintion
+  # list without evaluating the value.
+  lazyValueOf = type:
+    mkOptionType rec {
+      name = "lazyValueOf ${type.name}";
+      inherit (type) description descriptionClass emptyValue getSubOptions getSubModules;
+      check = isLazyValue;
+      merge = loc: defs:
+        assert assertMsg
+        (all (x: type.check x._lazyValue) defs)
+        "The option `${showOption loc}` is defined with a lazy value holding an invalid type";
+          types.mergeOneOption loc defs;
+      substSubModules = m: types.uniq (type.substSubModules m);
+      functor = (types.defaultFunctor name) // {wrapped = type;};
+      nestedTypes.elemType = type;
+    };
+
+  # Represents a value or lazy value of the given type that will
+  # automatically be coerced to the given type when merged.
+  lazyOf = type: types.coercedTo (lazyValueOf type) (x: x._lazyValue) type;
+in {
+  lib = recursiveUpdate prev.lib {
+    types = {
+      inherit
+        isLazyValue
+        lazyValue
+        lazyValueOf
+        lazyOf
+        ;
+    };
+  };
+}
--- a/modules/default.nix
+++ b/modules/default.nix
@ -1,7 +1,15 @@
-{
+{inputs, ...}: {
  imports = [
-    ./interface-naming.nix
+    inputs.microvm.nixosModules.host
+
    ./boot.nix
+    ./guests/default.nix
+    ./interface-naming.nix
    ./nginx.nix
+    ./node.nix
+  ];
+
+  nixpkgs.overlays = [
+    inputs.microvm.overlay
  ];
 }
--- a/modules/guests/common-guest-config.nix
+++ b/modules/guests/common-guest-config.nix
@ -0,0 +1,27 @@
+_guestName: guestCfg: {lib, ...}: let
+  inherit (lib) mkForce;
+in {
+  node.name = guestCfg.nodeName;
+  node.type = guestCfg.backend;
+
+  nix = {
+    settings.auto-optimise-store = mkForce false;
+    optimise.automatic = mkForce false;
+    gc.automatic = mkForce false;
+  };
+
+  systemd.network.networks."10-${guestCfg.networking.mainLinkName}" = {
+    matchConfig.Name = guestCfg.networking.mainLinkName;
+    DHCP = "yes";
+    # XXX: Do we really want this?
+    dhcpV4Config.UseDNS = false;
+    dhcpV6Config.UseDNS = false;
+    ipv6AcceptRAConfig.UseDNS = false;
+    networkConfig = {
+      IPv6PrivacyExtensions = "yes";
+      MulticastDNS = true;
+      IPv6AcceptRA = true;
+    };
+    linkConfig.RequiredForOnline = "routable";
+  };
+}
--- a/modules/guests/container.nix
+++ b/modules/guests/container.nix
@ -0,0 +1,64 @@
+guestName: guestCfg: {
+  config,
+  inputs,
+  lib,
+  pkgs,
+  ...
+}: let
+  inherit
+    (lib)
+    flip
+    mapAttrs'
+    nameValuePair
+    ;
+in {
+  ephemeral = true;
+  privateNetwork = true;
+  autoStart = guestCfg.autostart;
+  macvlans = ["${guestCfg.container.macvlan}:${guestCfg.networking.mainLinkName}"];
+  extraFlags = [
+    "--uuid=${builtins.substring 0 32 (builtins.hashString "sha256" guestName)}"
+  ];
+  bindMounts = flip mapAttrs' guestCfg.zfs (
+    _: zfsCfg:
+      nameValuePair zfsCfg.guestMountpoint {
+        hostPath = zfsCfg.hostMountpoint;
+        isReadOnly = false;
+      }
+  );
+  nixosConfiguration = inputs.nixpkgs.lib.nixosSystem {
+    specialArgs = guestCfg.extraSpecialArgs;
+    prefix = ["nodes" "${config.node.name}-${guestName}" "config"];
+    system = null;
+    modules =
+      [
+        {
+          boot.isContainer = true;
+          networking.useHostResolvConf = false;
+
+          # We cannot force the package set via nixpkgs.pkgs and
+          # inputs.nixpkgs.nixosModules.readOnlyPkgs, since some nixosModules
+          # like nixseparatedebuginfod depend on adding packages via nixpkgs.overlays.
+          # So we just mimic the options and overlays defined by the passed pkgs set.
+          nixpkgs.hostPlatform = config.nixpkgs.hostPlatform.system;
+          nixpkgs.overlays = pkgs.overlays;
+          nixpkgs.config = pkgs.config;
+
+          # Bind the /guest/* paths from above so impermancence doesn't complain.
+          # We bind-mount stuff from the host to itself, which is perfectly defined
+          # and not recursive. This allows us to have a fileSystems entry for each
+          # bindMount which other stuff can depend upon (impermanence adds dependencies
+          # to the state fs).
+          fileSystems = flip mapAttrs' guestCfg.zfs (_: zfsCfg:
+            nameValuePair zfsCfg.guestMountpoint {
+              neededForBoot = true;
+              fsType = "none";
+              device = zfsCfg.guestMountpoint;
+              options = ["bind"];
+            });
+        }
+        (import ./common-guest-config.nix guestName guestCfg)
+      ]
+      ++ guestCfg.modules;
+  };
+}
--- a/modules/guests/default.nix
+++ b/modules/guests/default.nix
@ -0,0 +1,281 @@
+{
+  config,
+  lib,
+  pkgs,
+  utils,
+  ...
+} @ attrs: let
+  inherit
+    (lib)
+    attrNames
+    attrValues
+    attrsToList
+    disko
+    escapeShellArg
+    flatten
+    flip
+    groupBy
+    hasPrefix
+    listToAttrs
+    literalExpression
+    makeBinPath
+    mapAttrs
+    mapAttrsToList
+    mergeToplevelConfigs
+    mkIf
+    mkMerge
+    mkOption
+    net
+    types
+    ;
+
+  # All available backends
+  backends = ["microvm" "container"];
+
+  guestsByBackend =
+    lib.genAttrs backends (_: {})
+    // mapAttrs (_: listToAttrs) (groupBy (x: x.value.backend) (attrsToList config.guests));
+
+  # List the necessary mount units for the given guest
+  fsMountUnitsFor = guestCfg:
+    map
+    (x: "${utils.escapeSystemdPath x.hostMountpoint}.mount")
+    (attrValues guestCfg.zfs);
+
+  # Configuration required on the host for a specific guest
+  defineGuest = _guestName: guestCfg: {
+    # Add the required datasets to the disko configuration of the machine
+    disko.devices.zpool = mkMerge (flip map (attrValues guestCfg.zfs) (zfsCfg: {
+      ${zfsCfg.pool}.datasets.${zfsCfg.dataset} = disko.filesystem zfsCfg.hostMountpoint;
+    }));
+
+    # Ensure that the zfs dataset exists before it is mounted.
+    systemd.services = mkMerge (flip map (attrValues guestCfg.zfs) (zfsCfg: let
+      fsMountUnit = "${utils.escapeSystemdPath zfsCfg.hostMountpoint}.mount";
+    in {
+      "zfs-ensure-${utils.escapeSystemdPath zfsCfg.hostMountpoint}" = {
+        wantedBy = [fsMountUnit];
+        before = [fsMountUnit];
+        after = [
+          "zfs-import-${utils.escapeSystemdPath zfsCfg.pool}.service"
+          "zfs-mount.target"
+        ];
+        unitConfig.DefaultDependencies = "no";
+        serviceConfig.Type = "oneshot";
+        script = let
+          poolDataset = "${zfsCfg.pool}/${zfsCfg.dataset}";
+          diskoDataset = config.disko.devices.zpool.${zfsCfg.pool}.datasets.${zfsCfg.dataset};
+        in ''
+          export PATH=${makeBinPath [pkgs.zfs]}":$PATH"
+          if ! zfs list -H -o type ${escapeShellArg poolDataset} &>/dev/null ; then
+            ${diskoDataset._create}
+          fi
+        '';
+      };
+    }));
+  };
+
+  defineMicrovm = guestName: guestCfg: {
+    # Ensure that the zfs dataset exists before it is mounted.
+    systemd.services."microvm@${guestName}" = {
+      requires = fsMountUnitsFor guestCfg;
+      after = fsMountUnitsFor guestCfg;
+    };
+
+    microvm.vms.${guestName} = import ./microvm.nix guestName guestCfg attrs;
+  };
+
+  defineContainer = guestName: guestCfg: {
+    # Ensure that the zfs dataset exists before it is mounted.
+    systemd.services."container@${guestName}" = {
+      requires = fsMountUnitsFor guestCfg;
+      after = fsMountUnitsFor guestCfg;
+      # Don't use the notify service type. Using exec will always consider containers
+      # started immediately and donesn't wait until the container is fully booted.
+      # Containers should behave like independent machines, and issues inside the container
+      # will unnecessarily lock up the service on the host otherwise.
+      # This causes issues on system activation or when containers take longer to start
+      # than TimeoutStartSec.
+      serviceConfig.Type = lib.mkForce "exec";
+    };
+
+    containers.${guestName} = import ./container.nix guestName guestCfg attrs;
+  };
+in {
+  imports = [
+    {
+      # This is opt-out, so we can't put this into the mkIf below
+      microvm.host.enable = guestsByBackend.microvm != {};
+    }
+  ];
+
+  options.node.type = mkOption {
+    type = types.enum (["host"] ++ backends);
+    description = "The type of this machine.";
+    default = "host";
+  };
+
+  options.containers = mkOption {
+    type = types.attrsOf (types.submodule (submod: {
+      options.nixosConfiguration = mkOption {
+        type = types.unspecified;
+        default = null;
+        description = "Set this to the result of a `nixosSystem` invocation to use it as the guest system. This will set the `path` option for you.";
+      };
+      config = mkIf (submod.config.nixosConfiguration != null) {
+        path = submod.config.nixosConfiguration.config.system.build.toplevel;
+      };
+    }));
+  };
+
+  options.guests = mkOption {
+    default = {};
+    description = "Defines the actual vms and handles the necessary base setup for them.";
+    type = types.attrsOf (types.submodule (submod: {
+      options = {
+        nodeName = mkOption {
+          type = types.str;
+          default = "${config.node.name}-${submod.config._module.args.name}";
+          description = ''
+            The name of the resulting node. By default this will be a compound name
+            of the host's name and the guest's name to avoid name clashes. Can be
+            overwritten to designate special names to specific guests.
+          '';
+        };
+
+        backend = mkOption {
+          type = types.enum backends;
+          description = ''
+            Determines how the guest will be hosted. You can currently choose
+            between microvm based deployment, or nixos containers.
+          '';
+        };
+
+        extraSpecialArgs = mkOption {
+          type = types.attrs;
+          default = {};
+          example = literalExpression "{ inherit inputs; }";
+          description = ''
+            Extra `specialArgs` passed to each guest system definition. This
+            option can be used to pass additional arguments to all modules.
+          '';
+        };
+
+        # Options for the microvm backend
+        microvm = {
+          system = mkOption {
+            type = types.str;
+            description = "The system that this microvm should use";
+          };
+
+          macvtap = mkOption {
+            type = types.str;
+            description = "The host interface to which the microvm should be attached via macvtap";
+          };
+
+          baseMac = mkOption {
+            type = types.net.mac;
+            description = "The base mac address from which the guest's mac will be derived. Only the second and third byte are used, so for 02:XX:YY:ZZ:ZZ:ZZ, this specifies XX and YY, while Zs are generated automatically. Not used if the mac is set directly.";
+            default = "02:01:27:00:00:00";
+          };
+
+          mac = mkOption {
+            type = types.net.mac;
+            description = "The MAC address for the guest's macvtap interface";
+            default = let
+              base = "02:${lib.substring 3 5 submod.config.microvm.baseMac}:00:00:00";
+            in
+              (net.mac.assignMacs base 24 [] (attrNames config.guests)).${submod.config._module.args.name};
+          };
+        };
+
+        # Options for the container backend
+        container = {
+          macvlan = mkOption {
+            type = types.str;
+            description = "The host interface to which the container should be attached";
+          };
+        };
+
+        networking.mainLinkName = mkOption {
+          type = types.str;
+          description = "The main ethernet link name inside of the guest. For containers, this cannot be named similar to an existing interface on the host.";
+          default =
+            if submod.config.backend == "microvm"
+            then submod.config.microvm.macvtap
+            else if submod.config.backend == "container"
+            then "mv-${submod.config.container.macvlan}"
+            else throw "Invalid backend";
+        };
+
+        zfs = mkOption {
+          description = "zfs datasets to mount into the guest";
+          default = {};
+          type = types.attrsOf (types.submodule (zfsSubmod: {
+            options = {
+              pool = mkOption {
+                type = types.str;
+                description = "The host's zfs pool on which the dataset resides";
+              };
+
+              dataset = mkOption {
+                type = types.str;
+                example = "safe/guests/mycontainer";
+                description = "The host's dataset that should be used for this mountpoint (will automatically be created, including parent datasets)";
+              };
+
+              hostMountpoint = mkOption {
+                type = types.path;
+                default = "/guests/${submod.config._module.args.name}${zfsSubmod.config.guestMountpoint}";
+                example = "/guests/mycontainer/persist";
+                description = "The host's mountpoint for the guest's dataset";
+              };
+
+              guestMountpoint = mkOption {
+                type = types.path;
+                default = zfsSubmod.config._module.args.name;
+                example = "/persist";
+                description = "The mountpoint inside the guest.";
+              };
+            };
+          }));
+        };
+
+        autostart = mkOption {
+          type = types.bool;
+          default = false;
+          description = "Whether this guest should be started automatically with the host";
+        };
+
+        modules = mkOption {
+          type = types.listOf types.unspecified;
+          default = [];
+          description = "Additional modules to load";
+        };
+      };
+    }));
+  };
+
+  config = mkIf (config.guests != {}) (
+    mkMerge [
+      {
+        systemd.tmpfiles.rules = [
+          "d /guests 0700 root root -"
+        ];
+
+        assertions = flatten (flip mapAttrsToList config.guests (
+          guestName: guestCfg:
+            flip mapAttrsToList guestCfg.zfs (
+              zfsName: zfsCfg: {
+                assertion = hasPrefix "/" zfsCfg.guestMountpoint;
+                message = "guest ${guestName}: zfs ${zfsName}: the guestMountpoint must be an absolute path.";
+              }
+            )
+        ));
+      }
+      (mergeToplevelConfigs ["disko" "systemd"] (mapAttrsToList defineGuest config.guests))
+      (mergeToplevelConfigs ["containers" "systemd"] (mapAttrsToList defineContainer guestsByBackend.container))
+      (mergeToplevelConfigs ["microvm" "systemd"] (mapAttrsToList defineMicrovm guestsByBackend.microvm))
+    ]
+  );
+}
--- a/modules/guests/microvm.nix
+++ b/modules/guests/microvm.nix
@ -0,0 +1,82 @@
+guestName: guestCfg: {
+  config,
+  inputs,
+  lib,
+  pkgs,
+  ...
+}: let
+  inherit
+    (lib)
+    flip
+    mapAttrsToList
+    mkDefault
+    mkForce
+    ;
+in {
+  specialArgs = guestCfg.extraSpecialArgs;
+  pkgs = inputs.self.pkgs.${guestCfg.microvm.system};
+  inherit (guestCfg) autostart;
+  config = {
+    imports =
+      guestCfg.modules
+      ++ [
+        (import ./common-guest-config.nix guestName guestCfg)
+        ({config, ...}: {
+          # Set early hostname too, so we can associate those logs to this host and don't get "localhost" entries in loki
+          boot.kernelParams = ["systemd.hostname=${config.networking.hostName}"];
+        })
+      ];
+
+    # TODO needed because of https://github.com/NixOS/nixpkgs/issues/102137
+    environment.noXlibs = mkForce false;
+    lib.microvm.mac = guestCfg.microvm.mac;
+
+    microvm = {
+      hypervisor = mkDefault "qemu";
+
+      # Give them some juice by default
+      mem = mkDefault (1024 + 2048);
+
+      # Add a writable store overlay, but since this is always ephemeral
+      # disable any store optimization from nix.
+      writableStoreOverlay = "/nix/.rw-store";
+
+      # MACVTAP bridge to the host's network
+      interfaces = [
+        {
+          type = "macvtap";
+          id = "vm-${guestName}";
+          inherit (guestCfg.microvm) mac;
+          macvtap = {
+            link = guestCfg.microvm.macvtap;
+            mode = "bridge";
+          };
+        }
+      ];
+
+      shares =
+        [
+          # Share the nix-store of the host
+          {
+            source = "/nix/store";
+            mountPoint = "/nix/.ro-store";
+            tag = "ro-store";
+            proto = "virtiofs";
+          }
+        ]
+        ++ flip mapAttrsToList guestCfg.zfs (
+          _: zfsCfg: {
+            source = zfsCfg.hostMountpoint;
+            mountPoint = zfsCfg.guestMountpoint;
+            tag = builtins.substring 0 16 (builtins.hashString "sha256" zfsCfg.hostMountpoint);
+            proto = "virtiofs";
+          }
+        );
+    };
+
+    networking.renameInterfacesByMac.${guestCfg.networking.mainLinkName} = guestCfg.microvm.mac;
+    systemd.network.networks."10-${guestCfg.networking.mainLinkName}".matchConfig = mkForce {
+      MACAddress = guestCfg.microvm.mac;
+    };
+  };
+}