frcc experiments (#1)

Co-authored-by: Autumn Liu <al63244@my.utexas.edu>
Co-authored-by: Floweynt <50057682+Floweynt@users.noreply.github.com>
Reviewed-on: #1
Co-authored-by: autumn <aut.tea7@gmail.com>

containers/server-n0/ssh.nix

@@ -61,16 +61,29 @@ let
       net ? { },
       resources ? { },
       traits ? [ ],
+      isolation ? { },
     }:
     mkContainer {
       inherit name;
       inherit net resources;
       svc.autoStart = true;
-      isolation.nixDaemon = nixDaemon;
+      isolation = {
+        inherit nixDaemon;
+      }
+      // isolation;
       traits = [ sshTrait ] ++ traits;
       packages = with pkgs; [ tmux ] ++ packages;
       miscProperties = self._sshpiperRegister routingInfo;
     };
+
+  kvmTrait = {
+    containerConfig =
+      { pkgs, ... }:
+      {
+        users.users.runner.extraGroups = [ "kvm" ];
+        environment.systemPackages = [ pkgs.qemu_kvm ];
+      };
+  };
 in
 {
   ssh-admin = mkSsh {
@@ -121,6 +134,25 @@ in
       mem = "16G";
       cpu = 65;
     };
+    traits = [ kvmTrait ];
+    isolation.allowedDevices = [
+      {
+        node = "/dev/kvm";
+        modifier = "rw";
+      }
+      {
+        node = "/dev/vhost-net";
+        modifier = "rw";
+      }
+      {
+        node = "/dev/net/tun";
+        modifier = "rwm";
+      }
+    ];
+    isolation.additionalCapabilities = [
+      "CAP_NET_ADMIN"
+      "CAP_MKNOD"
+    ];
     routingInfo = [
       {
         username = "autumn";

dsl.nix

@@ -228,6 +228,17 @@ rec {
     : systemd-nspawn user namespacing; see `--private-users=` in
       systemd-nspawn(1). Bind mounts always go through idmapped `--bind`.
 
+    `isolation.allowedDevices`
+    : List of `{ node; modifier; }` host devices to grant the container's
+      cgroup access to; see `containers.<name>.allowedDevices` in the NixOS
+      manual. nspawn bind-mounts a handful of devices (`/dev/kvm`,
+      `/dev/net/tun`, ...) into the container itself once the cgroup allows it -
+      this doesn't need a matching `resources.fs` bind mount.
+
+    `isolation.additionalCapabilities`
+    : Extra Linux capabilities granted to the container; see
+      `containers.<name>.additionalCapabilities` in the NixOS manual.
+
     `miscProperties`
     : Arbitrary data attached to the result. Used for cross-container discovery
       via `collectContainers`.
@@ -272,6 +283,8 @@ rec {
       bindMounts = resources.fs or { };
       allowsNixDaemon = isolation.nixDaemon or false;
       privateUsers = isolation.privateUsers or "pick";
+      allowedDevices = isolation.allowedDevices or [ ];
+      additionalCapabilities = isolation.additionalCapabilities or [ ];
 
       allowInternet = hasInternetAccess access;
       allBindMounts = builtins.foldl' (acc: t: acc // (t.resources.fs or { })) { } traits // bindMounts;
@@ -368,7 +381,12 @@ rec {
       kind = "nspawn";
       miscProperties = flattenMiscProp self miscProperties;
       nixosContainer = {
-        inherit autoStart privateUsers;
+        inherit
+          autoStart
+          privateUsers
+          allowedDevices
+          additionalCapabilities
+          ;
         bindMounts = { };
         extraFlags = bindMountExtraFlags;
         privateNetwork = true;