try fix again

.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():

configuration.nix

@@ -4,26 +4,12 @@
   lib,
   hostInfo,
   customPkgs,
-  flakeRev,
   ...
 }:
 let
   inherit (hostInfo) paths;
 in
 {
-  imports = [ ./modules/autodeploy.nix ];
-
-  services.server-autodeploy = {
-    enable = true;
-    giteaUrl = "https://git.floweytf.com/";
-    giteaRepo = "server/server-config";
-    tokenFile = "${hostInfo.paths.secrets}/gitea-autodeploy-token";
-    allowedSigners = [
-      "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIPaGpTX3EzfezVc2nl3MgsKVamYYP2h2VJaRny8pZkvi flowey@arch"
-    ];
-    stateDir = hostInfo.paths.config;
-  };
-
   assertions =
     let
       isMounted =
@@ -61,14 +47,6 @@ in
 
   time.timeZone = "UTC";
   system.stateVersion = "26.05";
-  system.nixos.label =
-    let
-      labelFile = ./.nixos-label;
-    in
-    if builtins.pathExists labelFile then
-      lib.removeSuffix "\n" (builtins.readFile labelFile)
-    else
-      flakeRev;
 
   nix = {
     settings = {

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"]

flake.nix

@@ -11,7 +11,6 @@
 
   outputs =
     {
-      self,
       nixpkgs,
       mtce,
       microvm,
@@ -22,8 +21,6 @@
       inherit (nixpkgs) lib;
       pkgs = import nixpkgs { inherit system; };
 
-      flakeRev = self.shortRev or self.dirtyShortRev or "unknown";
-
       customPkgs = {
         hopper-rs = pkgs.callPackage ./packages/hopper-rs.nix { };
         sshpiper = pkgs.callPackage ./packages/sshpiper.nix { };
@@ -32,7 +29,6 @@
         server-rebuild = pkgs.callPackage ./packages/server-rebuild.nix { };
         server-status = pkgs.callPackage ./packages/server-status.nix { };
         server-kexec = pkgs.callPackage ./packages/server-kexec.nix { };
-        server-autodeploy = pkgs.callPackage ./packages/server-autodeploy.nix { };
       };
 
       ci = import ./ci.nix { inherit lib pkgs; };
@@ -50,13 +46,12 @@
           server = "/persist/@containers";
           data = "/persist/@containers/data";
           secrets = "/persist/@secrets/";
-          config = "/persist/@config/";
         };
       };
 
       specialArgsFor = hostName: {
         hostInfo = hostInfoFor hostName;
-        inherit inputs customPkgs flakeRev;
+        inherit inputs customPkgs;
       };
 
       hostDefFor =
@@ -152,7 +147,6 @@
           server-rebuild
           server-status
           server-kexec
-          server-autodeploy
           ;
       };
     };

modules/autodeploy.nix

@@ -1,84 +0,0 @@
-{
-  config,
-  lib,
-  pkgs,
-  customPkgs,
-  ...
-}:
-let
-  cfg = config.services.server-autodeploy;
-in
-{
-  options.services.server-autodeploy = {
-    enable = lib.mkEnableOption "automatic deployment from signed git tags";
-
-    interval = lib.mkOption {
-      type = lib.types.str;
-      default = "15min";
-    };
-
-    branch = lib.mkOption {
-      type = lib.types.str;
-      default = "master";
-    };
-
-    giteaUrl = lib.mkOption {
-      type = lib.types.str;
-    };
-
-    giteaRepo = lib.mkOption {
-      type = lib.types.str;
-    };
-
-    tokenFile = lib.mkOption {
-      type = lib.types.str;
-    };
-
-    allowedSigners = lib.mkOption {
-      type = lib.types.listOf lib.types.str;
-      default = [ ];
-    };
-
-    stateDir = lib.mkOption {
-      type = lib.types.str;
-    };
-  };
-
-  config = lib.mkIf cfg.enable {
-    systemd.services.server-autodeploy = {
-      description = "Deploy latest signed tag";
-      after = [ "network-online.target" ];
-      wants = [ "network-online.target" ];
-
-      environment = {
-        GITEA_URL = cfg.giteaUrl;
-        GITEA_REPO = cfg.giteaRepo;
-        TOKEN_FILE = cfg.tokenFile;
-        ALLOWED_SIGNERS = builtins.toFile "signers" (lib.join "\n" cfg.allowedSigners);
-        DEPLOY_BRANCH = cfg.branch;
-        STATE_DIR = cfg.stateDir;
-        REPO_DIR = "/persist/@configuration";
-      };
-
-      serviceConfig = {
-        Type = "oneshot";
-        ExecStart = lib.getExe customPkgs.server-autodeploy;
-        TimeoutStartSec = "30min";
-      };
-
-      path = [
-        config.nix.package
-        pkgs.nixos-rebuild
-      ];
-    };
-
-    systemd.timers.server-autodeploy = {
-      wantedBy = [ "timers.target" ];
-      timerConfig = {
-        OnBootSec = "2min";
-        OnUnitActiveSec = cfg.interval;
-        RandomizedDelaySec = "2min";
-      };
-    };
-  };
-}

packages/server-autodeploy.nix

@@ -1,14 +0,0 @@
-{
-  writeShellApplication,
-  git,
-  curl,
-  jq,
-}:
-import ./mk-script.nix { inherit writeShellApplication; } {
-  name = "server-autodeploy";
-  runtimeInputs = [
-    git
-    curl
-    jq
-  ];
-}

scripts/server-autodeploy.sh

@@ -1,58 +0,0 @@
-REPO_DIR="${REPO_DIR:-/persist/@configuration}"
-STATE_DIR="${STATE_DIR:-/persist/@autodeploy}"
-BRANCH="origin/${DEPLOY_BRANCH:-master}"
-
-mkdir -p "$STATE_DIR"
-exec 9>"$STATE_DIR/lock"
-flock -n 9 || { echo "another deploy running"; exit 0; }
-
-report_status() {
-  [[ -f "$TOKEN_FILE" ]] || return 0
-  curl -sf -X POST \
-    -H "Authorization: token $(<"$TOKEN_FILE")" \
-    -H "Content-Type: application/json" \
-    -d "$(jq -n --arg s "$1" --arg d "$2" --arg c "deploy/$(hostname)" \
-      '{state:$s,description:$d,context:$c}')" \
-    "$GITEA_URL/api/v1/repos/$GITEA_REPO/statuses/$3" >/dev/null
-}
-
-cd "$REPO_DIR"
-
-export GIT_CONFIG_COUNT=2
-export GIT_CONFIG_KEY_0=gpg.format
-export GIT_CONFIG_VALUE_0=ssh
-export GIT_CONFIG_KEY_1=gpg.ssh.allowedSignersFile
-export GIT_CONFIG_VALUE_1="$ALLOWED_SIGNERS"
-
-git fetch origin --tags --prune-tags
-
-latest=$(git describe --abbrev=0 "$BRANCH" 2>/dev/null) || {
-  echo "no tags on $BRANCH"
-  exit 0
-}
-
-current=$(cat "$STATE_DIR/tag" 2>/dev/null) || true
-[[ "$latest" != "$current" ]] || { echo "already at $latest"; exit 0; }
-
-echo "new tag: $latest (current: ${current:-none})"
-
-git tag -v "$latest" || {
-  sha=$(git rev-list -1 "$latest")
-  report_status "failure" "tag signature verification failed" "$sha"
-  exit 1
-}
-
-sha=$(git rev-list -1 "$latest")
-report_status "pending" "deploying $latest" "$sha"
-
-# build from remote, pinned to the verified rev - no local checkout needed
-flake_url="git+${GITEA_URL}/${GITEA_REPO}?ref=refs/tags/${latest}&rev=${sha}#$(hostname)"
-
-if nixos-rebuild switch --flake "$flake_url"; then
-  echo "$latest" > "$STATE_DIR/tag"
-  report_status "success" "deployed $latest" "$sha"
-  echo "deployed $latest"
-else
-  report_status "failure" "nixos-rebuild failed" "$sha"
-  exit 1
-fi

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"