flake.lock: update inputs

.github/workflows/diff-pr.yml

@@ -17,10 +17,18 @@ jobs:
     steps:
       - uses: actions/checkout@v4
         with:
+          ref: ${{ github.event.pull_request.head.sha }}
           fetch-depth: 0
 
+      - name: merge base branch
+        run: |
+          git config user.email "ci@localhost"
+          git config user.name "CI"
+          git fetch origin ${{ github.event.pull_request.base.ref }}
+          git merge --no-edit origin/${{ github.event.pull_request.base.ref }}
+
       - name: worktree base for diff
-        run: git worktree add "$RUNNER_TEMP/base" "${{ github.event.pull_request.base.sha }}"
+        run: git worktree add "$RUNNER_TEMP/base" "origin/${{ github.event.pull_request.base.ref }}"
 
       - name: build closures and render diff
         run: |

.github/workflows/flake-update.yml

@@ -5,21 +5,19 @@ on:
     - cron: "17 4 * * *"
   workflow_dispatch:
 
-permissions:
-  contents: write
-  pull-requests: write
-
 jobs:
   update:
     runs-on: nix
     steps:
       - uses: actions/checkout@v4
+        with:
+          persist-credentials: false
 
       - run: nix flake update -L
 
       - name: open/update PR
         env:
-          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          GH_TOKEN: ${{ secrets.BOT_TOKEN }}
           API: ${{ github.api_url }}
           REPO: ${{ github.repository }}
         run: |

ci.py

@@ -8,6 +8,7 @@ import re
 import subprocess
 import subprocess_tee
 import sys
+import urllib.error
 import urllib.request
 from collections import Counter
 from pathlib import Path
@@ -26,6 +27,10 @@ def git(*args, check=True):
     return run("git", *args, check=check, capture_output=True)
 
 
+class GiteaError(Exception):
+    pass
+
+
 def gitea(method, path, body=None):
     url = f'{os.environ["API"]}/repos/{os.environ["REPO"]}{path}'
     data = json.dumps(body).encode() if body is not None else None
@@ -37,7 +42,11 @@ def gitea(method, path, body=None):
         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()}")
+        try:
+            detail = json.loads(e.read()).get("message", e.reason)
+        except (json.JSONDecodeError, UnicodeDecodeError):
+            detail = e.reason
+        raise GiteaError(f"{method} {path}: {e.code} {detail}") from None
     return json.loads(raw) if raw else None
 
 
@@ -466,16 +475,24 @@ def open_pr(args):
 
     host = os.environ["API"].removesuffix("/api/v1")
     auth = base64.b64encode(f'token:{os.environ["GH_TOKEN"]}'.encode()).decode()
-    git("-c", f"http.extraHeader=Authorization: Basic {auth}",
-        "push", "-f", f'{host}/{os.environ["REPO"]}.git', f"HEAD:{args.branch}")
+    push = git("-c", f"http.extraHeader=Authorization: Basic {auth}",
+               "push", "-f", f'{host}/{os.environ["REPO"]}.git', f"HEAD:{args.branch}",
+               check=False)
+    if push.returncode != 0:
+        raise RuntimeError(f"git push failed:\n{push.stderr}")
 
-    owner = os.environ["REPO"].split("/")[0]
-    if gitea("GET", f"/pulls?state=open&head={owner}:{args.branch}"):
+    pulls = gitea("GET", f"/pulls?state=open&base_branch={args.base}")
+    existing = next((pr for pr in pulls if pr["head"]["ref"] == args.branch), None)
+    if existing:
+        print(f"PR #{existing['number']} already open, pushed update")
         return
     labels = [label_id(args.label)] if args.label else []
-    gitea("POST", "/pulls",
+    pr = gitea("POST", "/pulls",
                {"title": args.title, "head": args.branch, "base": args.base,
                 "body": Path(args.body_file).read_text(), "labels": labels})
+    if not pr or "number" not in pr:
+        raise GiteaError(f"POST /pulls: unexpected response: {json.dumps(pr)}")
+    print(f"opened PR #{pr['number']}")
 
 
 def main():

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

@@ -221,17 +221,6 @@ 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.
 
@@ -239,6 +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`.
@@ -281,15 +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=")
@@ -297,10 +298,6 @@ 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=${d}") devices;
       inherit (deriveIdentity ifPrefix.nspawn name)
         iface
         mac
@@ -379,15 +376,19 @@ rec {
         hostAddr
         bindMountHostPaths
         includedInTests
-        devices
         ;
       bindMounts = allBindMounts;
       kind = "nspawn";
       miscProperties = flattenMiscProp self miscProperties;
       nixosContainer = {
-        inherit autoStart privateUsers;
+        inherit
+          autoStart
+          privateUsers
+          allowedDevices
+          additionalCapabilities
+          ;
         bindMounts = { };
-        extraFlags = bindMountExtraFlags ++ deviceFlags;
+        extraFlags = bindMountExtraFlags;
         privateNetwork = true;
         extraVeths = linkVeths;
         config =
@@ -520,7 +521,6 @@ rec {
         ;
       inherit needsGateway;
       bindMountHostPaths = map (s: s.source) shares;
-      devices = [ ];
       kind = "vm";
       miscProperties = flattenMiscProp self miscProperties;
       microvmConfig = {
@@ -957,7 +957,6 @@ rec {
           startLimitBurst = 5;
           startLimitIntervalSec = 120;
           unitConfig.RequiresMountsFor = c.bindMountHostPaths;
-          serviceConfig.DeviceAllow = map (d: "${d} rw") c.devices;
         };
 
       # the host end of a link, for both kinds. nixos-containers only configures
@@ -1317,7 +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"]

flake.lock

@@ -26,11 +26,11 @@
         "spectrum": "spectrum"
       },
       "locked": {
-        "lastModified": 1786300091,
-        "narHash": "sha256-4UFJOVGpaYtHW5yasSv80MaJxTBYdk2zyf3jhKtt0wA=",
+        "lastModified": 1789506833,
+        "narHash": "sha256-4L8SOCF84de1irB72Ka9VHcu9wrAJ9Amuey4lU9oh4Y=",
         "owner": "microvm-nix",
         "repo": "microvm.nix",
-        "rev": "71beea0076cd46dafcee97a5a2e7d00cbba5bd2f",
+        "rev": "187b0a390ee054028106a674e7b01b1cb940cbba",
         "type": "github"
       },
       "original": {
@@ -63,11 +63,11 @@
     },
     "nixpkgs": {
       "locked": {
-        "lastModified": 1786384358,
-        "narHash": "sha256-RzPPiWeUtuvymnpuEWsdtzli5w4kjZs49FqEs3/1u+I=",
+        "lastModified": 1789546076,
+        "narHash": "sha256-zVxLZiSnmaaPLwnhj7pwmqe3axBg/C6nG5JZsJMh2g4=",
         "owner": "NixOS",
         "repo": "nixpkgs",
-        "rev": "2fcb964de67fcf60b43471c55d5d99e61a9ccb5a",
+        "rev": "b1b875982b17dabde9b4a37f3e229e74913e6db3",
         "type": "github"
       },
       "original": {
@@ -87,11 +87,11 @@
     "spectrum": {
       "flake": false,
       "locked": {
-        "lastModified": 1785761586,
-        "narHash": "sha256-MWOMVqJJERwjQGgRIv8d6rlEBqbLM3VZWxHkNKIIZNw=",
+        "lastModified": 1786032750,
+        "narHash": "sha256-+Ba/kIn5BjvV7PN8bWOo7RoqgTY/Yv2M7qeMf3wS2Ag=",
         "ref": "refs/heads/main",
-        "rev": "a7762d6f54b40560dd5255ce902e6e6a5d980fe9",
-        "revCount": 1416,
+        "rev": "77d2389874bb092b8dbf05ecc5b1030372e80e98",
+        "revCount": 1417,
         "type": "git",
         "url": "https://spectrum-os.org/git/spectrum"
       },

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}"
@@ -329,40 +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:
-            # 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"