actually bind /dev/kvm into the container, and test it initializes

DeviceAllow= only grants cgroup permission to open a device node - it
doesn't create one, and nspawn can't mknod a device inside a user
namespace. The node needs an explicit bind from the host as well, so
both are required together.

Also add a test that actually boots qemu-system-x86_64 -enable-kvm
(no disk, killed by timeout) and checks for "Permission denied" /
"Could not access KVM" in its output, rather than just checking the
node opens - a wrong-gid or unmapped node still opens but fails the
ioctls qemu needs.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dzy7tEKLCFFXD96LmR8we3

.github/workflows/diff-pr.yml

@@ -13,26 +13,21 @@ concurrency:
 
 jobs:
   dix:
-    runs-on: ubuntu-latest
+    runs-on: nix
     steps:
       - uses: actions/checkout@v4
         with:
           fetch-depth: 0
 
-      - uses: cachix/install-nix-action@v31
-        with:
-          extra_nix_config: |
-            experimental-features = nix-command flakes
-
       - name: worktree base for diff
         run: git worktree add "$RUNNER_TEMP/base" "${{ github.event.pull_request.base.sha }}"
 
       - name: build closures and render diff
         run: |
-          nix run .#ci -- diff "$RUNNER_TEMP/base" . \
+          nix run -L .#ci -- diff "$RUNNER_TEMP/base" . \
             "### System closure diff vs \`${{ github.event.pull_request.base.ref }}\`" \
             > "$RUNNER_TEMP/comment.md"
-          nix run .#ci -- inspect-diff "$RUNNER_TEMP/base" . \
+          nix run -L .#ci -- inspect-diff "$RUNNER_TEMP/base" . \
             "### Inspection diff vs \`${{ github.event.pull_request.base.ref }}\`" \
             >> "$RUNNER_TEMP/comment.md"
 
@@ -42,4 +37,4 @@ jobs:
           API: ${{ github.api_url }}
           REPO: ${{ github.repository }}
           PR: ${{ github.event.pull_request.number }}
-        run: nix run .#ci -- post-comment "<!-- dix-diff -->" "$RUNNER_TEMP/comment.md"
+        run: nix run -L .#ci -- post-comment "<!-- dix-diff -->" "$RUNNER_TEMP/comment.md"

.github/workflows/diff-tag.yml

@@ -9,17 +9,12 @@ permissions:
 
 jobs:
   dix:
-    runs-on: ubuntu-latest
+    runs-on: nix
     steps:
       - uses: actions/checkout@v4
         with:
           fetch-depth: 0
 
-      - uses: cachix/install-nix-action@v31
-        with:
-          extra_nix_config: |
-            experimental-features = nix-command flakes
-
       - name: diff latest tag against HEAD
         run: |
           tag=$(git tag --sort=-creatordate | head -n1)
@@ -30,8 +25,8 @@ jobs:
           git worktree add "$RUNNER_TEMP/old" "$tag"
           head=$(git rev-parse --short HEAD)
           {
-            nix run .#ci -- diff "$RUNNER_TEMP/old" . \
+            nix run -L .#ci -- diff "$RUNNER_TEMP/old" . \
               "# System closure diff: \`$tag\` -> \`$head\`"
-            nix run .#ci -- inspect-diff "$RUNNER_TEMP/old" . \
+            nix run -L .#ci -- inspect-diff "$RUNNER_TEMP/old" . \
               "# Inspection diff: \`$tag\` -> \`$head\`"
           } | tee -a "$GITHUB_STEP_SUMMARY"

.github/workflows/flake-check.yml

@@ -14,13 +14,8 @@ concurrency:
 
 jobs:
   flake-check:
-    runs-on: ubuntu-latest
+    runs-on: nix
     steps:
       - uses: actions/checkout@v4
 
-      - uses: cachix/install-nix-action@v31
-        with:
-          extra_nix_config: |
-            experimental-features = nix-command flakes
-
       - run: nix flake check -L

.github/workflows/flake-update.yml

@@ -11,16 +11,11 @@ permissions:
 
 jobs:
   update:
-    runs-on: ubuntu-latest
+    runs-on: nix
     steps:
       - uses: actions/checkout@v4
 
-      - uses: cachix/install-nix-action@v31
-        with:
-          extra_nix_config: |
-            experimental-features = nix-command flakes
-
-      - run: nix flake update
+      - run: nix flake update -L
 
       - name: open/update PR
         env:
@@ -29,5 +24,5 @@ jobs:
           REPO: ${{ github.repository }}
         run: |
           printf 'Automated `nix flake update`.\n' > "$RUNNER_TEMP/body.md"
-          nix run .#ci -- open-pr \
+          nix run -L .#ci -- open-pr \
             flake-update "$GITHUB_REF_NAME" "flake.lock: update inputs" "$RUNNER_TEMP/body.md" automated

.github/workflows/lint.yml

@@ -15,27 +15,22 @@ concurrency:
 
 jobs:
   lint:
-    runs-on: ubuntu-latest
+    runs-on: nix
     steps:
       - uses: actions/checkout@v4
 
-      - uses: cachix/install-nix-action@v31
-        with:
-          extra_nix_config: |
-            experimental-features = nix-command flakes
-
-      - name: pick reporter
-        run: |
-          if [ "${{ github.event_name }}" = pull_request ]; then
-            echo "REPORTER=github-pr-review" >> "$GITHUB_ENV"
-          else
-            echo "REPORTER=local" >> "$GITHUB_ENV"
-          fi
-
       - name: run linters
-        run: nix run .#ci -- lint . "$RUNNER_TEMP/findings"
+        run: nix run -L .#ci -- lint . "$RUNNER_TEMP/findings"
 
       - name: report findings
+        if: always()
         env:
-          REVIEWDOG_GITHUB_API_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        run: nix run .#ci -- lint-report "$REPORTER" "$RUNNER_TEMP/findings"
+          REVIEWDOG_GITEA_API_TOKEN: ${{ secrets.GITEA_TOKEN }}
+          GITEA_ADDRESS: ${{ github.server_url }}
+        run: |
+          if [ "${{ github.event_name }}" = pull_request ]; then
+            nix run -L .#ci -- lint-report gitea-pr-review "$RUNNER_TEMP/findings"
+          else
+            nix run -L .#ci -- lint-summary "$RUNNER_TEMP/findings" \
+              | tee -a "$GITHUB_STEP_SUMMARY"
+          fi

.github/workflows/vulnix.yml

@@ -5,24 +5,15 @@ on:
     - cron: "0 6 * * 1"
   workflow_dispatch:
 
-permissions:
-  contents: read
-  issues: write
-
 jobs:
   scan:
-    runs-on: ubuntu-latest
+    runs-on: nix
     steps:
       - uses: actions/checkout@v4
 
-      - uses: cachix/install-nix-action@v31
-        with:
-          extra_nix_config: |
-            experimental-features = nix-command flakes
-
       - name: scan host closures
         id: scan
-        run: echo "count=$(nix run .#ci -- vulnix . "$RUNNER_TEMP/issue.md")" >> "$GITHUB_OUTPUT"
+        run: echo "count=$(nix run -L .#ci -- vulnix . "$RUNNER_TEMP/issue.md")" >> "$GITHUB_OUTPUT"
 
       - name: file/update issue
         if: steps.scan.outputs.count != '0'
@@ -30,7 +21,7 @@ jobs:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           API: ${{ github.api_url }}
           REPO: ${{ github.repository }}
-        run: nix run .#ci -- upsert-issue vulnix "vulnix: known vulnerabilities in system closure" "$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'
@@ -38,4 +29,4 @@ jobs:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           API: ${{ github.api_url }}
           REPO: ${{ github.repository }}
-        run: nix run .#ci -- close-issue vulnix
+        run: nix run -L .#ci -- close-issue vulnix

ci.nix

@@ -1,9 +1,17 @@
 { 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

@@ -6,6 +6,7 @@ import json
 import os
 import re
 import subprocess
+import subprocess_tee
 import sys
 import urllib.request
 from collections import Counter
@@ -17,7 +18,8 @@ def run(*args, **kw):
 
 
 def nix(*args):
-    return run("nix", "--extra-experimental-features", "nix-command flakes", *args, check=True, capture_output=True).stdout
+    # 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
 
 
 def git(*args, check=True):
@@ -31,13 +33,16 @@ def gitea(method, path, body=None):
     req.add_header("Authorization", f'token {os.environ["GH_TOKEN"]}')
     if data is not None:
         req.add_header("Content-Type", "application/json")
+    try:
         with urllib.request.urlopen(req) as resp:
             raw = resp.read()
+    except urllib.error.HTTPError as e:
+        sys.exit(f"gitea {method} {path}: {e.code} {e.reason}\n{e.read().decode()}")
     return json.loads(raw) if raw else None
 
 
 def find_open_issue(label):
-    issues = gitea("GET", f"/issues?state=open&labels={label}")
+    issues = gitea("GET", f"/issues?state=open&type=issues&labels={label}")
     return issues[0] if issues else None
 
 
@@ -234,7 +239,7 @@ def format_vulnix(vulns):
     for v in vulns:
         for cve in v["affected_by"]:
             score = v.get("cvssv3_basescore", {}).get(cve)
-            desc = v.get("description", {}).get(cve, "")
+            desc = v.get("description", {}).get(cve, "").replace("\n", " ").replace("|", "∣")
             if len(desc) > 120:
                 desc = desc[:117] + "..."
             rows.append((v["pname"], v["version"], cve, score, desc))
@@ -362,10 +367,63 @@ def lint_report(args):
 
     codes = [reviewdog(f"{args.dir}/{name}", flags)
              for name, flags in LINTERS
-             if os.path.getsize(f"{args.dir}/{name}") > 0]
+             if os.path.isfile(f"{args.dir}/{name}") and os.path.getsize(f"{args.dir}/{name}") > 0]
     sys.exit(1 if any(codes) else 0)
 
 
+def has_findings(path):
+    return os.path.isfile(path) and os.path.getsize(path) > 0
+
+
+def fmt_rdjsonl(path):
+    lines = []
+    with open(path) as f:
+        for raw in f:
+            rec = json.loads(raw)
+            loc = rec["location"]
+            r = loc.get("range", {}).get("start")
+            pos = f"{loc['path']}:{r['line']}:{r['column']}" if r else loc["path"]
+            lines.append(f"- `{pos}`: {rec['message']}")
+    return "\n".join(lines)
+
+
+def fmt_errfmt(path):
+    lines = []
+    with open(path) as f:
+        for raw in f:
+            parts = raw.strip().split(">", 1)
+            if len(parts) == 2:
+                lines.append(f"- `{parts[0]}:{parts[1].split(':', 3)[-1].strip()}`" if ":" in parts[1] else f"- `{raw.strip()}`")
+            elif raw.strip():
+                lines.append(f"- `{raw.strip()}`")
+    return "\n".join(lines)
+
+
+def lint_summary(args):
+    sections = []
+    d = args.dir
+
+    if has_findings(f"{d}/nixfmt.diff"):
+        with open(f"{d}/nixfmt.diff") as f:
+            sections.append(f"## nixfmt\n\n```diff\n{f.read().rstrip()}\n```")
+
+    if has_findings(f"{d}/statix.errfmt"):
+        sections.append(f"## statix\n\n{fmt_errfmt(f'{d}/statix.errfmt')}")
+
+    if has_findings(f"{d}/deadnix.rdjsonl"):
+        sections.append(f"## deadnix\n\n{fmt_rdjsonl(f'{d}/deadnix.rdjsonl')}")
+
+    if has_findings(f"{d}/cycles.rdjsonl"):
+        sections.append(f"## systemd ordering cycles\n\n{fmt_rdjsonl(f'{d}/cycles.rdjsonl')}")
+
+    if sections:
+        print("\n\n".join(["# Lint findings"] + sections))
+        sys.exit(1)
+    else:
+        print("No lint findings.")
+        sys.exit(0)
+
+
 def post_comment(args):
     body = args.marker + "\n" + Path(args.body_file).read_text()
     pr = os.environ["PR"]
@@ -438,6 +496,7 @@ def main():
     command("vulnix", vulnix, "ref", "body_out")
     command("lint", lint, "target", "out")
     command("lint-report", lint_report, "reporter", "dir")
+    command("lint-summary", lint_summary, "dir")
     command("post-comment", post_comment, "marker", "body_file")
     command("upsert-issue", upsert_issue, "label", "title", "body_file")
     command("close-issue", close_issue, "label")

containers/common/overseer.nix

@@ -381,6 +381,6 @@ let
         };
       };
     };
-  hostAddr = c.hostAddr;
+  inherit (c) hostAddr;
 in
 c

containers/server-n0/git.nix

@@ -8,6 +8,9 @@ let
   inherit (dsl) mkContainer mkVm;
   inherit (hostInfo.paths) data secrets;
 
+  pubSshPort = 23;
+  intSshPort = 23 + 8000;
+
   ciRunnerData = "${data}/web/git-ci-runner";
   ciRunnerSecrets = "${secrets}/ci-runner";
 in
@@ -17,8 +20,8 @@ in
     net.access = [ { to = "@internet"; } ];
     net.forward.tcp = [
       {
-        host = 23;
-        container = 8023;
+        host = pubSshPort;
+        container = intSshPort;
       }
     ];
     svc.autoStart = true;
@@ -38,7 +41,8 @@ in
             ROOT_URL = "https://git.floweytf.com/";
             START_SSH_SERVER = true;
             BUILTIN_SSH_SERVER_USER = "git";
-            SSH_PORT = 8023;
+            SSH_PORT = pubSshPort;
+            SSH_LISTEN_PORT = intSshPort;
           };
           service.DISABLE_REGISTRATION = true;
         };
@@ -51,8 +55,10 @@ in
     net.expose.tcp = [ 3000 ];
   };
 
-  ci-runner = mkVm {
+  ci-runner =
+    mkVm {
       name = "ci-runner";
+      includedInTests = false;
       net.access = [
         {
           to = "git";
@@ -62,7 +68,7 @@ in
       ];
       svc.autoStart = true;
       resources.cpu = 4;
-    resources.mem = "8G";
+      resources.mem = "32G";
       resources.fs = [
         {
           proto = "virtiofs";
@@ -70,6 +76,12 @@ in
           source = "${ciRunnerData}/nix-rw-store";
           mountPoint = "/nix/.rw-store";
         }
+        {
+          proto = "virtiofs";
+          tag = "nix-db";
+          source = "${ciRunnerData}/nix-db";
+          mountPoint = "/nix/var/nix/db";
+        }
         {
           proto = "virtiofs";
           tag = "docker-data";
@@ -80,7 +92,7 @@ in
           proto = "virtiofs";
           tag = "runner-state";
           source = "${ciRunnerData}/runner-state";
-        mountPoint = "/var/lib/gitea-runner";
+          mountPoint = "/var/lib/private/gitea-runner";
         }
         {
           proto = "virtiofs";
@@ -89,6 +101,12 @@ in
           mountPoint = "/run/ci-runner-secrets";
           readOnly = true;
         }
+        {
+          proto = "virtiofs";
+          tag = "tmp";
+          source = "/tmp";
+          mountPoint = "/tmp";
+        }
       ];
       guestConfig =
         {
@@ -100,20 +118,24 @@ in
         let
           runnerInstanceName = "ci-runner";
 
-        ciJobImage = pkgs.dockerTools.buildLayeredImage {
-          name = "ci-job-image";
-          tag = "latest";
-          contents = with pkgs; [
+          ciJobContents = with pkgs; [
             bashInteractive
             coreutils
             gitMinimal
             nixVersions.latest
             nodejs
             cacert
+            (writeTextDir "etc/nix/nix.conf" "experimental-features = nix-command flakes\n")
           ];
+
+          ciJobImage = pkgs.dockerTools.buildLayeredImage {
+            name = "ci-job-image";
+            tag = "latest";
+            contents = ciJobContents;
             config = {
               Cmd = [ "/bin/sh" ];
               Env = [
+                "PATH=${lib.makeBinPath ciJobContents}"
                 "SSL_CERT_FILE=${pkgs.cacert}/etc/ssl/certs/ca-bundle.crt"
                 "NIX_SSL_CERT_FILE=${pkgs.cacert}/etc/ssl/certs/ca-bundle.crt"
                 "NIX_REMOTE=daemon"
@@ -125,6 +147,10 @@ in
           boot.kernelModules = [ "kvm-intel" ];
           boot.extraModprobeConfig = "options kvm_intel nested=1";
 
+          # volume-mounted into CI containers; must be in VM's store so
+          # image symlinks (/bin/node etc.) resolve through the mount
+          system.extraDependencies = ciJobContents;
+
           microvm.writableStoreOverlay = "/nix/.rw-store";
 
           nix.settings = {
@@ -133,16 +159,35 @@ in
               "flakes"
             ];
             max-jobs = "auto";
+            substituters = [ "https://microvm.cachix.org" ];
+            trusted-public-keys = [ "microvm.cachix.org-1:oXnBc6hRE3eX5rSYdRyMYXnfzcCxC7yKPTbZXALsqys=" ];
           };
 
-        nix.gc = {
-          automatic = true;
-          dates = "daily";
-          options = "--delete-older-than 14d";
+          nix.settings = {
+            min-free = 5 * 1024 * 1024 * 1024;
+            max-free = 10 * 1024 * 1024 * 1024;
           };
           # nix.optimise.automatic = true; - doesn't work with microvm.writableStoreOverlay
 
+          services.openssh = {
+            enable = true;
+            settings = {
+              PermitRootLogin = "yes";
+              PermitEmptyPasswords = "yes";
+            };
+          };
+          users.users.root.initialHashedPassword = "";
+          services.journald.extraConfig = "ForwardToConsole=yes";
+
+          virtualisation.containerd.enable = true;
+          virtualisation.containerd.settings.plugins."io.containerd.snapshotter.v1.overlayfs".mount_options =
+            [ "userxattr" ];
+
           virtualisation.docker.enable = true;
+          virtualisation.docker.extraOptions = "--containerd /run/containerd/containerd.sock";
+          virtualisation.docker.daemon.settings.exec-opts = [ "native.cgroupdriver=cgroupfs" ];
+          systemd.services.docker.wants = [ "containerd.service" ];
+          systemd.services.docker.after = [ "containerd.service" ];
 
           systemd.services.load-ci-job-image = {
             description = "Load the CI job Docker image";
@@ -174,5 +219,15 @@ in
             };
           };
         };
+    }
+    // {
+      hostConfig = {
+        systemd.tmpfiles.rules = [
+          "d ${ciRunnerData}/nix-rw-store 0755 server server -"
+          "d ${ciRunnerData}/nix-db 0755 server server -"
+          "d ${ciRunnerData}/docker 0755 server server -"
+          "d ${ciRunnerData}/runner-state 0755 server server -"
+        ];
+      };
     };
 }

containers/server-n0/nginx.nix

@@ -210,6 +210,6 @@ let
     };
     inherit (setup) containerConfig;
   };
-  hostAddr = c.hostAddr;
+  inherit (c) hostAddr;
 in
 c

containers/server-n0/ssh.nix

@@ -61,12 +61,16 @@ 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;
@@ -113,14 +117,28 @@ 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

@@ -221,6 +221,17 @@ rec {
     `resources.fs`
     : Bind mounts, keyed by mount point.
 
+    `resources.devices`
+    : Host device nodes to expose, as paths. Bound into the container at the
+      same path and added to the unit's `DeviceAllow=` - the cgroup
+      permission alone doesn't get you a working node (nothing can `mknod` a
+      device inside a user namespace), and the bind alone leaves the cgroup
+      denying `open()`, so both are needed. The bind is unconditional, so
+      the device must actually exist on whatever host runs the container.
+      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.
 
@@ -270,6 +281,7 @@ 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";
 
@@ -277,7 +289,7 @@ rec {
       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 != "no";
+      useIdmap = privateUsers == "pick" || (builtins.isInt privateUsers && privateUsers > 0);
       bindMountExtraFlags = lib.mapAttrsToList (
         mountPoint: bm:
         (if bm.isReadOnly then "--bind-ro=" else "--bind=")
@@ -285,6 +297,10 @@ rec {
         + lib.optionalString useIdmap ":idmap"
       ) allBindMounts;
       bindMountHostPaths = map (bm: bm.hostPath) (lib.attrValues allBindMounts);
+      # DeviceAllow= alone only grants cgroup permission - nspawn can't
+      # mknod a device node inside a user namespace, so the node still
+      # needs an explicit bind from the host.
+      deviceFlags = map (d: "--bind=${d}") devices;
       inherit (deriveIdentity ifPrefix.nspawn name)
         iface
         mac
@@ -363,6 +379,7 @@ rec {
         hostAddr
         bindMountHostPaths
         includedInTests
+        devices
         ;
       bindMounts = allBindMounts;
       kind = "nspawn";
@@ -370,7 +387,7 @@ rec {
       nixosContainer = {
         inherit autoStart privateUsers;
         bindMounts = { };
-        extraFlags = bindMountExtraFlags;
+        extraFlags = bindMountExtraFlags ++ deviceFlags;
         privateNetwork = true;
         extraVeths = linkVeths;
         config =
@@ -503,6 +520,7 @@ rec {
         ;
       inherit needsGateway;
       bindMountHostPaths = map (s: s.source) shares;
+      devices = [ ];
       kind = "vm";
       miscProperties = flattenMiscProp self miscProperties;
       microvmConfig = {
@@ -553,6 +571,11 @@ rec {
               Destination = "0.0.0.0/0";
               Gateway = hostAddr;
             };
+            networkConfig.DNS = [ hostAddr ];
+            networkConfig.Domains = [
+              "~."
+              "containers"
+            ];
             linkConfig.RequiredForOnline = if allowInternet then "routable" else "degraded";
           };
           networking.enableIPv6 = false;
@@ -934,6 +957,7 @@ rec {
           startLimitBurst = 5;
           startLimitIntervalSec = 120;
           unitConfig.RequiresMountsFor = c.bindMountHostPaths;
+          serviceConfig.DeviceAllow = map (d: "${d} rw") c.devices;
         };
 
       # the host end of a link, for both kinds. nixos-containers only configures
@@ -1293,6 +1317,7 @@ 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"]

flake.nix

@@ -129,11 +129,10 @@
     {
       nixosConfigurations = lib.mapAttrs (_: host: host.hostCfg) allHosts;
 
-      inspect = lib.mapAttrs
-        (name: host:
-          pkgs.writeText "inspect-${name}.json"
-            (builtins.toJSON (introspect.hostReport { inherit host; })))
-        allHosts;
+      inspect = lib.mapAttrs (
+        name: host:
+        pkgs.writeText "inspect-${name}.json" (builtins.toJSON (introspect.hostReport { inherit host; }))
+      ) allHosts;
 
       packages.${system} = customPkgs // {
         inherit ci;

tests/server-n0/integration.py

@@ -27,6 +27,7 @@ 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 = ""
@@ -55,6 +56,11 @@ 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}"
@@ -205,12 +211,7 @@ with subtest("(isolation): edges are one-way"):
     server.fail(ct_connect("git", ADDR["nginx"], 443, timeout=2))
 
 with subtest("(isolation): host access allowlist"):
-    server.succeed(
-        net_cmd(
-            "overseer",
-            f"{CURL} -sf --max-time 5 http://{HOST_ADDR['overseer']}:{HOST_EXPORTER_PORT}/metrics >&2",
-        )
-    )
+    server.succeed(ct_connect("overseer", HOST_ADDR["overseer"], HOST_EXPORTER_PORT))
     for port in (HOST_EXPORTER_PORT, HOST_CADVISOR_PORT):
         server.fail(ct_connect("git", HOST_ADDR["git"], port, timeout=2))
         server.fail(ct_connect("git", HOST_ADDR["overseer"], port, timeout=2))
@@ -328,3 +329,40 @@ 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:
+            # opened, not stat'd: without DeviceAllow= the node is still visible
+            server.succeed(ct_cmd(name, f"exec 3<> {dev}"))
+
+
+with subtest("(functional): /dev/kvm actually initializes, not just opens"):
+    # a wrong-gid or unmapped node still opens but fails the ioctls qemu
+    # needs, so boot with no disk (never completes) and kill it - the
+    # failure mode this catches is qemu refusing to start at all.
+    for name, devices in DEVICES.items():
+        if "/dev/kvm" not in devices:
+            continue
+        out = server.succeed(
+            ct_cmd(
+                name,
+                f"{TIMEOUT} 3 qemu-system-x86_64 -enable-kvm -m 256 -nographic -no-reboot "
+                ">/tmp/qemu.out 2>&1; cat /tmp/qemu.out",
+            )
+        )
+        assert "Permission denied" not in out, f"{name}: qemu couldn't use /dev/kvm:\n{out}"
+        assert "Could not access KVM" not in out, f"{name}: qemu couldn't use /dev/kvm:\n{out}"
+
+
+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"