better cache management

.github/workflows/vulnix.yml

@@ -25,7 +25,7 @@ jobs:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           API: ${{ github.api_url }}
           REPO: ${{ github.repository }}
-        run: 'nix run -L .#ci -- upsert-issue vulnix "vulnix: known vulnerabilities in system closure" "$RUNNER_TEMP/issue.md"'
+        run: nix run -L .#ci -- upsert-issue vulnix "vulnix: known vulnerabilities in system closure" "$RUNNER_TEMP/issue.md"
 
       - name: close resolved issue
         if: steps.scan.outputs.count == '0'

ci.nix

@@ -1,17 +1,10 @@
 { lib, pkgs }:
-let
-  # vulnix's bundled nix-2.34 broke its own --experimental-features detection
-  nixConf = pkgs.writeText "ci-nix.conf" "experimental-features = nix-command flakes";
-in
 # tools must track ci.py's subprocess calls; nix stays ambient (uses the
 # runner's own nix, not a pinned one).
 pkgs.writers.writePython3Bin "ci" {
   flakeIgnore = [ "E501" ];
   libraries = [ pkgs.python3Packages.subprocess-tee ];
   makeWrapperArgs = [
-    "--set-default"
-    "NIX_USER_CONF_FILES"
-    (toString nixConf)
     "--prefix"
     "PATH"
     ":"

ci.py

@@ -18,8 +18,7 @@ def run(*args, **kw):
 
 
 def nix(*args):
-    # tee to stderr - stdout leaks into GITHUB_OUTPUT $() captures
-    return subprocess_tee.run(("nix", "--extra-experimental-features", "nix-command flakes", "-L", *args), check=True, text=True, stdout=sys.stderr).stdout
+    return subprocess_tee.run(("nix", "--extra-experimental-features", "nix-command flakes", "-L", *args), check=True, text=True).stdout
 
 
 def git(*args, check=True):

containers/server-n0/ssh.nix

@@ -61,21 +61,16 @@ let
       net ? { },
       resources ? { },
       traits ? [ ],
-      isolation ? { },
     }:
     mkContainer {
       inherit name;
       inherit net resources;
       svc.autoStart = true;
-      isolation = {
-        inherit nixDaemon;
-      }
-      // isolation;
+      isolation.nixDaemon = nixDaemon;
       traits = [ sshTrait ] ++ traits;
       packages = with pkgs; [ tmux ] ++ packages;
       miscProperties = self._sshpiperRegister routingInfo;
     };
-
 in
 {
   ssh-admin = mkSsh {
@@ -118,28 +113,14 @@ 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 = [
-      {
-        containerConfig =
-          { config, ... }:
-          {
-            users.groups.kvm.gid = config.ids.gids.kvm;
-            users.users.runner.extraGroups = [ "kvm" ];
-          };
-      }
-    ];
     routingInfo = [
       {
         username = "autumn";

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.
 
@@ -279,7 +270,6 @@ 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";
 
@@ -373,7 +363,6 @@ rec {
         hostAddr
         bindMountHostPaths
         includedInTests
-        devices
         ;
       bindMounts = allBindMounts;
       kind = "nspawn";
@@ -514,7 +503,6 @@ rec {
         ;
       inherit needsGateway;
       bindMountHostPaths = map (s: s.source) shares;
-      devices = [ ];
       kind = "vm";
       miscProperties = flattenMiscProp self miscProperties;
       microvmConfig = {
@@ -951,7 +939,6 @@ 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 +1298,6 @@ 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}"
@@ -329,27 +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:
-            if server.execute(f"test -e {dev}")[0] != 0:
-                # host lacks the device (e.g. /dev/kvm without nested
-                # virtualization, as in this test's own VM) - nspawn skips
-                # populating it in-container, so there's nothing to check.
-                continue
-            # opened, not stat'd: without DeviceAllow= the node is still visible
-            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"