fix failing passes

containers/server-n0/mail.nix

@@ -146,8 +146,6 @@ let
 in
 mkContainer {
   inherit name;
-  includedInTests = false; # TODO: fixme, clearly mail is flakey
-
   net.access = [ { to = "@internet"; } ];
   svc.autoStart = true;
   net.forward.tcp = [

containers/server-n0/ssh.nix

@@ -75,6 +75,15 @@ let
       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 {
@@ -117,27 +126,32 @@ in
   ssh-generic = mkSsh {
     name = "ssh-generic";
     net.access = [ { to = "@internet"; } ];
-    # /dev/kvm keeps its host ownership, so the kvm gid must not be shifted.
-    isolation.privateUsers = "identity";
     resources = {
       fs."/home/runner/data" = {
         hostPath = "${data}/ssh/generic";
         isReadOnly = false;
       };
-      devices = [ "/dev/kvm" ];
       mem = "16G";
       cpu = 65;
     };
-    packages = with pkgs; [ qemu_kvm ];
-    traits = [
+    traits = [ kvmTrait ];
+    isolation.allowedDevices = [
       {
-        containerConfig =
-          { config, ... }:
-          {
-            users.groups.kvm.gid = config.ids.gids.kvm;
-            users.users.runner.extraGroups = [ "kvm" ];
-          };
+        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 = [
       {

dsl.nix

@@ -221,15 +221,6 @@ rec {
     `resources.fs`
     : Bind mounts, keyed by mount point.
 
-    `resources.devices`
-    : Host device nodes to expose, as paths, added to the unit's
-      `DeviceAllow=`. nspawn populates a handful of privileged devices
-      (`/dev/kvm`, `/dev/net/tun`, ...) into the container's own `/dev` once
-      the cgroup allows it, and silently skips ones the host doesn't have -
-      no `resources.fs` bind mount needed. Nodes keep their host ownership,
-      so a consumer inside needs `isolation.privateUsers = "identity"` for
-      the owning gid to mean anything.
-
     `isolation.nixDaemon`
     : Whether the Nix daemon and CLI are available inside the container.
 
@@ -237,6 +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`.
@@ -279,15 +281,16 @@ rec {
       cpuQuota = "${toString (cpu * 100)}%";
       memoryMax = resources.mem or "512M";
       bindMounts = resources.fs or { };
-      devices = resources.devices 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;
       userModules = map mkUser (lib.concatMap (t: t.users or [ ]) traits ++ users);
       allPorts = ports ++ map (p: (normalizePort p).container) hostPorts;
-      useIdmap = privateUsers == "pick" || (builtins.isInt privateUsers && privateUsers > 0);
+      useIdmap = privateUsers != "no";
       bindMountExtraFlags = lib.mapAttrsToList (
         mountPoint: bm:
         (if bm.isReadOnly then "--bind-ro=" else "--bind=")
@@ -373,13 +376,17 @@ rec {
         hostAddr
         bindMountHostPaths
         includedInTests
-        devices
         ;
       bindMounts = allBindMounts;
       kind = "nspawn";
       miscProperties = flattenMiscProp self miscProperties;
       nixosContainer = {
-        inherit autoStart privateUsers;
+        inherit
+          autoStart
+          privateUsers
+          allowedDevices
+          additionalCapabilities
+          ;
         bindMounts = { };
         extraFlags = bindMountExtraFlags;
         privateNetwork = true;
@@ -514,7 +521,6 @@ rec {
         ;
       inherit needsGateway;
       bindMountHostPaths = map (s: s.source) shares;
-      devices = [ ];
       kind = "vm";
       miscProperties = flattenMiscProp self miscProperties;
       microvmConfig = {
@@ -951,7 +957,6 @@ rec {
           startLimitBurst = 5;
           startLimitIntervalSec = 120;
           unitConfig.RequiresMountsFor = c.bindMountHostPaths;
-          serviceConfig.DeviceAllow = map (d: "${d} rw") c.devices;
         };
 
       # the host end of a link, for both kinds. nixos-containers only configures
@@ -1311,7 +1316,6 @@ rec {
         HOST_ADDR = ${pyDict pyStr (lib.mapAttrs (_: c: c.hostAddr) autoStartTestContainers)}
         IFACE = ${pyDict pyStr (lib.mapAttrs (_: c: c.iface) autoStartTestContainers)}
         KIND = ${pyDict pyStr (lib.mapAttrs (_: c: c.kind) autoStartTestContainers)}
-        DEVICES = ${pyDict pyList (lib.mapAttrs (_: c: c.devices) autoStartTestContainers)}
         # only nspawn guests can be entered with nsenter; a vm has no host-side
         # netns, so every probe that runs *inside* a guest is limited to these
         enterable = [c for c in included if KIND[c] == "nspawn"]

tests/server-n0/integration.py

@@ -27,7 +27,6 @@ if TYPE_CHECKING:
     HOST_ADDR: dict[str, str] = {}
     IFACE: dict[str, str] = {}
     KIND: dict[str, str] = {}
-    DEVICES: dict[str, list[str]] = {}
     EDGES: list[tuple[str, str, list[int], list[int]]] = []
     SSH_KEY: str = ""
     SSH_OPTS: str = ""
@@ -56,11 +55,6 @@ def container_pid(name):
     return server.succeed(f"machinectl show {name} -p Leader --value").strip()
 
 
-def ct_cmd(name, cmd):
-    pid = container_pid(name)
-    return f"{NSENTER} --mount=/proc/{pid}/ns/mnt --root=/proc/{pid}/root -- {SH} -c '{cmd}'"
-
-
 def net_cmd(name, cmd):
     pid = container_pid(name)
     return f"{NSENTER} --net=/proc/{pid}/ns/net -- {cmd}"
@@ -153,8 +147,6 @@ with subtest("(functional): web services respond"):
     for name, port in [("devenv", 8080), ("maven", 8080), ("git", 3000)]:
         server.succeed(net_cmd(name, f"{CURL} -sf --max-time 5 -o /dev/null http://{ADDR[name]}:{port}/ >&2"))
 
-# TODO: re-enable mail
-"""
 with subtest("(functional): mail accepts connections"):
     smtp_probe = net_cmd("mail", f"{SH} -c 'exec 3<>/dev/tcp/{ADDR['mail']}/25 && head -1 <&3'")
     server.wait_until_succeeds(f"{smtp_probe} | grep -q '^220'", timeout=60)
@@ -169,7 +161,6 @@ with subtest("(functional): mail accepts connections"):
         return greeting.strip().startswith("* OK")
 
     retry(imap_ready, timeout_seconds=60)
-"""
 
 with subtest("(functional): nginx proxies to backend containers"):
     server.succeed(
@@ -200,7 +191,7 @@ with subtest("(isolation): undeclared container edges are refused"):
         ("devenv", "git", 3000),
         ("git", "devenv", 8080),
         ("maven", "git", 3000),
-        # ("nginx", "mail", 25),
+        ("nginx", "mail", 25),
         ("ssh-botm", "ssh-admin", 22),
         ("overseer", "git", 3000),
     ]:
@@ -253,7 +244,7 @@ with subtest("(isolation): renaming the interface changes nothing"):
     rename(iface, "evil0")
     try:
         server.succeed(ct_connect("nginx", ADDR["git"], 3000))
-        # server.fail(ct_connect("nginx", ADDR["mail"], 25, timeout=2))
+        server.fail(ct_connect("nginx", ADDR["mail"], 25, timeout=2))
     finally:
         rename("evil0", iface)
 
@@ -332,21 +323,3 @@ with subtest("(functional): published ports follow the container's lifetime"):
     server.succeed("systemctl start container@nginx")
     server.wait_for_unit("container@nginx.service")
     internet.wait_until_succeeds(ext_connect(80), timeout=30)
-
-with subtest("(functional): declared devices are usable in-container"):
-    for name, devices in DEVICES.items():
-        for dev in devices:
-            server.succeed(ct_cmd(name, f"exec 3<> {dev}"))
-
-
-with subtest("(isolation): devices reach only the container that declared them"):
-    for name in enterable:
-        for dev in {d for ds in DEVICES.values() for d in ds} - set(DEVICES.get(name, [])):
-            server.fail(ct_cmd(name, f"test -e {dev}"))
-
-
-with subtest("(isolation): identity uid mapping keeps the user namespace"):
-    for name in (n for n, ds in DEVICES.items() if ds):
-        host_ns = server.succeed("readlink /proc/1/ns/user").strip()
-        ct_ns = server.succeed(f"readlink /proc/{container_pid(name)}/ns/user").strip()
-        assert host_ns != ct_ns, f"{name} shares the host user namespace"