release: 0.4.4

Signed-off-by: Ophestra <cat@gensokyo.uk>

.clang-format

@@ -0,0 +1,2 @@
+ColumnLimit: 0
+IndentWidth: 4

.gitea/workflows/test.yml

@@ -2,7 +2,6 @@ name: Test
 
 on:
   - push
-  - pull_request
 
 jobs:
   hakurei:
@@ -73,20 +72,20 @@ jobs:
           path: result/*
           retention-days: 1
 
-  hpkg:
-    name: Hpkg
+  sharefs:
+    name: ShareFS
     runs-on: nix
     steps:
       - name: Checkout
         uses: actions/checkout@v4
 
       - name: Run NixOS test
-        run: nix build --out-link "result" --print-out-paths --print-build-logs .#checks.x86_64-linux.hpkg
+        run: nix build --out-link "result" --print-out-paths --print-build-logs .#checks.x86_64-linux.sharefs
 
       - name: Upload test output
         uses: actions/upload-artifact@v3
         with:
-          name: "hpkg-vm-output"
+          name: "sharefs-vm-output"
           path: result/*
           retention-days: 1
 
@@ -97,7 +96,7 @@ jobs:
       - race
       - sandbox
       - sandbox-race
-      - hpkg
+      - sharefs
     runs-on: nix
     steps:
       - name: Checkout

.github/workflows/README

@@ -1,5 +0,0 @@
-DO NOT ADD NEW ACTIONS HERE
-
-This port is solely for releasing to the github mirror and serves no purpose during development.
-All development happens at https://git.gensokyo.uk/security/hakurei. If you wish to contribute,
-request for an account on git.gensokyo.uk.

.github/workflows/release.yml

@@ -1,46 +0,0 @@
-name: Release
-
-on:
-  push:
-    tags:
-      - 'v*'
-
-jobs:
-  release:
-    name: Create release
-    runs-on: ubuntu-latest
-
-    permissions:
-      packages: write
-      contents: write
-
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-
-      - name: Install Nix
-        uses: nixbuild/nix-quick-install-action@v32
-        with:
-          nix_conf: |
-            keep-env-derivations = true
-            keep-outputs = true
-
-      - name: Restore and cache Nix store
-        uses: nix-community/cache-nix-action@v6
-        with:
-          primary-key: build-${{ runner.os }}-${{ hashFiles('**/*.nix') }}
-          restore-prefixes-first-match: build-${{ runner.os }}-
-          gc-max-store-size-linux: 1G
-          purge: true
-          purge-prefixes: build-${{ runner.os }}-
-          purge-created: 60
-          purge-primary-key: never
-
-      - name: Build for release
-        run: nix build --print-out-paths --print-build-logs .#dist
-
-      - name: Release
-        uses: softprops/action-gh-release@v2
-        with:
-          files: |-
-            result/hakurei-**

.github/workflows/test.yml

@@ -1,48 +0,0 @@
-name: Test
-
-on:
-  - push
-
-jobs:
-  dist:
-    name: Create distribution
-    runs-on: ubuntu-latest
-    permissions:
-      actions: write
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-
-      - name: Install Nix
-        uses: nixbuild/nix-quick-install-action@v32
-        with:
-          nix_conf: |
-            keep-env-derivations = true
-            keep-outputs = true
-
-      - name: Restore and cache Nix store
-        uses: nix-community/cache-nix-action@v6
-        with:
-          primary-key: build-${{ runner.os }}-${{ hashFiles('**/*.nix') }}
-          restore-prefixes-first-match: build-${{ runner.os }}-
-          gc-max-store-size-linux: 1G
-          purge: true
-          purge-prefixes: build-${{ runner.os }}-
-          purge-created: 60
-          purge-primary-key: never
-
-      - name: Build for test
-        id: build-test
-        run: >-
-          export HAKUREI_REV="$(git rev-parse --short HEAD)" &&
-          sed -i.old 's/version = /version = "0.0.0-'$HAKUREI_REV'"; # version = /' package.nix &&
-          nix build --print-out-paths --print-build-logs .#dist &&
-          mv package.nix.old package.nix &&
-          echo "rev=$HAKUREI_REV" >> $GITHUB_OUTPUT
-
-      - name: Upload test build
-        uses: actions/upload-artifact@v4
-        with:
-          name: "hakurei-${{ steps.build-test.outputs.rev }}"
-          path: result/*
-          retention-days: 1

.gitignore

@@ -1,35 +1,18 @@
-# Binaries for programs and plugins
-*.exe
-*.exe~
-*.dll
-*.so
-*.dylib
-*.pkg
-/hakurei
-
-# Test binary, built with `go test -c`
+# produced by tools and text editors
+*.qcow2
 *.test
-
-# Output of the go coverage tool, specifically when used with LiteIDE
 *.out
-
-# Dependency directories (remove the comment below to include it)
-# vendor/
-
-# Go workspace file
-go.work
-go.work.sum
-
-# env file
-.env
 .idea
 .vscode
 
 # go generate
 /cmd/hakurei/LICENSE
+/cmd/mbf/internal/pkgserver/ui/static
+/internal/pkg/internal/testtool/testtool
+/internal/rosa/hakurei_current.tar.gz
 
-# release
-/dist/hakurei-*
+# cmd/dist default destination
+/dist
 
-# interactive nixos vm
-nixos.qcow2
+# local packages
+/internal/rosa/package/local

README.md

@@ -1,5 +1,5 @@
 <p align="center">
-  <a href="https://git.gensokyo.uk/security/hakurei">
+  <a href="https://git.gensokyo.uk/rosa/hakurei">
     <picture>
       <img src="https://basement.gensokyo.uk/images/yukari1.png" width="200px" alt="Yukari">
     </picture>
@@ -8,171 +8,58 @@
 
 <p align="center">
   <a href="https://pkg.go.dev/hakurei.app"><img src="https://pkg.go.dev/badge/hakurei.app.svg" alt="Go Reference" /></a>
-  <a href="https://git.gensokyo.uk/security/hakurei/actions"><img src="https://git.gensokyo.uk/security/hakurei/actions/workflows/test.yml/badge.svg?branch=staging&style=flat-square" alt="Gitea Workflow Status" /></a>
+  <a href="https://git.gensokyo.uk/rosa/hakurei/actions"><img src="https://git.gensokyo.uk/rosa/hakurei/actions/workflows/test.yml/badge.svg?branch=staging&style=flat-square" alt="Gitea Workflow Status" /></a>
   <br/>
-  <a href="https://git.gensokyo.uk/security/hakurei/releases"><img src="https://img.shields.io/gitea/v/release/security/hakurei?gitea_url=https%3A%2F%2Fgit.gensokyo.uk&color=purple" alt="Release" /></a>
+  <a href="https://git.gensokyo.uk/rosa/hakurei/releases"><img src="https://img.shields.io/gitea/v/release/rosa/hakurei?gitea_url=https%3A%2F%2Fgit.gensokyo.uk&color=purple" alt="Release" /></a>
   <a href="https://goreportcard.com/report/hakurei.app"><img src="https://goreportcard.com/badge/hakurei.app" alt="Go Report Card" /></a>
   <a href="https://hakurei.app"><img src="https://img.shields.io/website?url=https%3A%2F%2Fhakurei.app" alt="Website" /></a>
 </p>
 
-Hakurei is a tool for running sandboxed graphical applications as dedicated subordinate users on the Linux kernel.
-It implements the application container of [planterette (WIP)](https://git.gensokyo.uk/security/planterette),
-a self-contained Android-like package manager with modern security features.
+Hakurei is a tool for running sandboxed desktop applications as dedicated
+subordinate users on the Linux kernel. It implements the application container
+of [planterette (WIP)](https://git.gensokyo.uk/rosa/planterette), a
+self-contained Android-like package manager with modern security features.
 
-## NixOS Module usage
+Interaction with hakurei happens entirely through structures described by
+package [hst](https://pkg.go.dev/hakurei.app/hst). No native API is available
+due to internal details of uid isolation.
 
-The NixOS module currently requires home-manager to configure subordinate users. Full module documentation can be found [here](options.md).
+## Notable Packages
 
-To use the module, import it into your configuration with
+Package [container](https://pkg.go.dev/hakurei.app/container) is general purpose
+container tooling. It is used by the hakurei shim process running as the target
+subordinate user to set up the application container. It has a single dependency,
+[libseccomp](https://github.com/seccomp/libseccomp), to create BPF programs
+for the [system call filter](https://www.kernel.org/doc/html/latest/userspace-api/seccomp_filter.html).
 
-```nix
-{
-  inputs = {
-    nixpkgs.url = "github:NixOS/nixpkgs/nixos-24.11";
+Package [internal/pkg](https://pkg.go.dev/hakurei.app/internal/pkg) provides
+infrastructure for hermetic builds. This replaces the legacy nix-based testing
+framework and serves as the build system of Rosa OS, currently developed under
+package [internal/rosa](https://pkg.go.dev/hakurei.app/internal/rosa).
 
-    hakurei = {
-      url = "git+https://git.gensokyo.uk/security/hakurei";
+## Dependencies
 
-      # Optional but recommended to limit the size of your system closure.
-      inputs.nixpkgs.follows = "nixpkgs";
-    };
-  };
+`container` depends on:
 
-  outputs = { self, nixpkgs, hakurei, ... }:
-  {
-    nixosConfigurations.hakurei = nixpkgs.lib.nixosSystem {
-      system = "x86_64-linux";
-      modules = [
-        hakurei.nixosModules.hakurei
-      ];
-    };
-  };
-}
-```
+- [libseccomp](https://github.com/seccomp/libseccomp) to generate BPF programs.
 
-This adds the `environment.hakurei` option:
+`cmd/hakurei` depends on:
 
-```nix
-{ pkgs, ... }:
+- [acl](https://savannah.nongnu.org/projects/acl/) to export sockets to
+  subordinate users.
+- [wayland](https://gitlab.freedesktop.org/wayland/wayland) to set up
+  [security-context-v1](https://wayland.app/protocols/security-context-v1).
+- [xcb](https://xcb.freedesktop.org/) to grant and revoke subordinate users
+  access to the X server.
 
-{
-  environment.hakurei = {
-    enable = true;
-    stateDir = "/var/lib/hakurei";
-    users = {
-      alice = 0;
-      nixos = 10;
-    };
+`cmd/sharefs` depends on:
 
-    commonPaths = [
-      {
-        src = "/sdcard";
-        write = true;
-      }
-    ];
+- [fuse](https://github.com/libfuse/libfuse) to implement the filesystem.
 
-    extraHomeConfig = {
-      home.stateVersion = "23.05";
-    };
+New dependencies will generally not be added. Patches adding new dependencies
+are very likely to be rejected.
 
-    apps = {
-      "org.chromium.Chromium" = {
-        name = "chromium";
-        identity = 1;
-        packages = [ pkgs.chromium ];
-        userns = true;
-        mapRealUid = true;
-        dbus = {
-          system = {
-            filter = true;
-            talk = [
-              "org.bluez"
-              "org.freedesktop.Avahi"
-              "org.freedesktop.UPower"
-            ];
-          };
-          session =
-            f:
-            f {
-              talk = [
-                "org.freedesktop.FileManager1"
-                "org.freedesktop.Notifications"
-                "org.freedesktop.ScreenSaver"
-                "org.freedesktop.secrets"
-                "org.kde.kwalletd5"
-                "org.kde.kwalletd6"
-              ];
-              own = [
-                "org.chromium.Chromium.*"
-                "org.mpris.MediaPlayer2.org.chromium.Chromium.*"
-                "org.mpris.MediaPlayer2.chromium.*"
-              ];
-              call = { };
-              broadcast = { };
-            };
-        };
-      };
+## NixOS Module (deprecated)
 
-      "org.claws_mail.Claws-Mail" = {
-        name = "claws-mail";
-        identity = 2;
-        packages = [ pkgs.claws-mail ];
-        gpu = false;
-        capability.pulse = false;
-      };
-
-      "org.weechat" = {
-        name = "weechat";
-        identity = 3;
-        shareUid = true;
-        packages = [ pkgs.weechat ];
-        capability = {
-          wayland = false;
-          x11 = false;
-          dbus = true;
-          pulse = false;
-        };
-      };
-
-      "dev.vencord.Vesktop" = {
-        name = "discord";
-        identity = 3;
-        shareUid = true;
-        packages = [ pkgs.vesktop ];
-        share = pkgs.vesktop;
-        command = "vesktop --ozone-platform-hint=wayland";
-        userns = true;
-        mapRealUid = true;
-        capability.x11 = true;
-        dbus = {
-          session =
-            f:
-            f {
-              talk = [ "org.kde.StatusNotifierWatcher" ];
-              own = [ ];
-              call = { };
-              broadcast = { };
-            };
-          system.filter = true;
-        };
-      };
-
-      "io.looking-glass" = {
-        name = "looking-glass-client";
-        identity = 4;
-        useCommonPaths = false;
-        groups = [ "plugdev" ];
-        extraPaths = [
-          {
-            src = "/dev/shm/looking-glass";
-            write = true;
-          }
-        ];
-        extraConfig = {
-          programs.looking-glass-client.enable = true;
-        };
-      };
-    };
-  };
-}
-```
+The NixOS module is in maintenance mode and will be removed once planterette is
+feature-complete. Full module documentation can be found [here](options.md).

all.sh

@@ -0,0 +1,3 @@
+#!/bin/sh -e
+
+HAKUREI_DIST_MAKE='' exec "$(dirname -- "$0")/cmd/dist/dist.sh"

check/absolute.go

@@ -0,0 +1,137 @@
+// Package check provides types yielding values checked to meet a condition.
+package check
+
+import (
+	"encoding"
+	"errors"
+	"fmt"
+	"path/filepath"
+	"slices"
+	"strings"
+	"syscall"
+	"unique"
+)
+
+// AbsoluteError is returned by [NewAbs] and holds the invalid pathname.
+type AbsoluteError string
+
+func (e AbsoluteError) Error() string {
+	return fmt.Sprintf("path %q is not absolute", string(e))
+}
+
+func (e AbsoluteError) Is(target error) bool {
+	ce, ok := errors.AsType[AbsoluteError](target)
+	if !ok {
+		return errors.Is(target, syscall.EINVAL)
+	}
+	return e == ce
+}
+
+// Absolute holds a pathname checked to be absolute.
+type Absolute struct{ pathname unique.Handle[string] }
+
+var (
+	_ fmt.GoStringer = new(Absolute)
+
+	_ encoding.TextAppender    = new(Absolute)
+	_ encoding.TextMarshaler   = new(Absolute)
+	_ encoding.TextUnmarshaler = new(Absolute)
+
+	_ encoding.BinaryAppender    = new(Absolute)
+	_ encoding.BinaryMarshaler   = new(Absolute)
+	_ encoding.BinaryUnmarshaler = new(Absolute)
+)
+
+func (a *Absolute) GoString() string {
+	return fmt.Sprintf("check.MustAbs(%q)", a.String())
+}
+
+// ok returns whether [Absolute] is not the zero value.
+func (a *Absolute) ok() bool { return a != nil && *a != (Absolute{}) }
+
+// unsafeAbs returns [check.Absolute] on any string value.
+func unsafeAbs(pathname string) *Absolute {
+	return &Absolute{unique.Make(pathname)}
+}
+
+// String returns the checked pathname.
+func (a *Absolute) String() string {
+	if !a.ok() {
+		panic("attempted use of zero Absolute")
+	}
+	return a.pathname.Value()
+}
+
+// Handle returns the underlying [unique.Handle].
+func (a *Absolute) Handle() unique.Handle[string] {
+	return a.pathname
+}
+
+// Is efficiently compares the underlying pathname.
+func (a *Absolute) Is(v *Absolute) bool {
+	if a == nil && v == nil {
+		return true
+	}
+	return a.ok() && v.ok() && a.pathname == v.pathname
+}
+
+// NewAbs checks pathname and returns a new [Absolute] if pathname is absolute.
+func NewAbs(pathname string) (*Absolute, error) {
+	if !filepath.IsAbs(pathname) {
+		return nil, AbsoluteError(pathname)
+	}
+	return unsafeAbs(pathname), nil
+}
+
+// MustAbs calls [NewAbs] and panics on error.
+func MustAbs(pathname string) *Absolute {
+	if a, err := NewAbs(pathname); err != nil {
+		panic(err)
+	} else {
+		return a
+	}
+}
+
+// Append calls [filepath.Join] with [Absolute] as the first element.
+func (a *Absolute) Append(elem ...string) *Absolute {
+	return unsafeAbs(filepath.Join(append([]string{a.String()}, elem...)...))
+}
+
+// Dir calls [filepath.Dir] with [Absolute] as its argument.
+func (a *Absolute) Dir() *Absolute { return unsafeAbs(filepath.Dir(a.String())) }
+
+// AppendText appends the checked pathname.
+func (a *Absolute) AppendText(data []byte) ([]byte, error) {
+	return append(data, a.String()...), nil
+}
+
+// MarshalText returns the checked pathname.
+func (a *Absolute) MarshalText() ([]byte, error) { return a.AppendText(nil) }
+
+// UnmarshalText stores data if it represents an absolute pathname.
+func (a *Absolute) UnmarshalText(data []byte) error {
+	pathname := string(data)
+	if !filepath.IsAbs(pathname) {
+		return AbsoluteError(pathname)
+	}
+	a.pathname = unique.Make(pathname)
+	return nil
+}
+
+func (a *Absolute) AppendBinary(data []byte) ([]byte, error) { return a.AppendText(data) }
+func (a *Absolute) MarshalBinary() ([]byte, error)           { return a.MarshalText() }
+func (a *Absolute) UnmarshalBinary(data []byte) error        { return a.UnmarshalText(data) }
+
+// SortAbs calls [slices.SortFunc] for a slice of [Absolute].
+func SortAbs(x []*Absolute) {
+	slices.SortFunc(x, func(a, b *Absolute) int {
+		return strings.Compare(a.String(), b.String())
+	})
+}
+
+// CompactAbs calls [slices.CompactFunc] for a slice of [Absolute].
+func CompactAbs(s []*Absolute) []*Absolute {
+	return slices.CompactFunc(s, func(a *Absolute, b *Absolute) bool {
+		return a.Is(b)
+	})
+}

check/absolute_test.go

@@ -1,4 +1,4 @@
-package container
+package check_test
 
 import (
 	"bytes"
@@ -9,9 +9,19 @@ import (
 	"strings"
 	"syscall"
 	"testing"
+	_ "unsafe" // for go:linkname
+
+	. "hakurei.app/check"
 )
 
+// unsafeAbs returns check.Absolute on any string value.
+//
+//go:linkname unsafeAbs hakurei.app/check.unsafeAbs
+func unsafeAbs(pathname string) *Absolute
+
 func TestAbsoluteError(t *testing.T) {
+	t.Parallel()
+
 	testCases := []struct {
 		name string
 
@@ -21,8 +31,8 @@ func TestAbsoluteError(t *testing.T) {
 	}{
 		{"EINVAL", new(AbsoluteError), syscall.EINVAL, true},
 		{"not EINVAL", new(AbsoluteError), syscall.EBADE, false},
-		{"ne val", new(AbsoluteError), &AbsoluteError{"etc"}, false},
-		{"equals", &AbsoluteError{"etc"}, &AbsoluteError{"etc"}, true},
+		{"ne val", new(AbsoluteError), AbsoluteError("etc"), false},
+		{"equals", AbsoluteError("etc"), AbsoluteError("etc"), true},
 	}
 
 	for _, tc := range testCases {
@@ -32,14 +42,18 @@ func TestAbsoluteError(t *testing.T) {
 	}
 
 	t.Run("string", func(t *testing.T) {
+		t.Parallel()
+
 		want := `path "etc" is not absolute`
-		if got := (&AbsoluteError{"etc"}).Error(); got != want {
+		if got := (AbsoluteError("etc")).Error(); got != want {
 			t.Errorf("Error: %q, want %q", got, want)
 		}
 	})
 }
 
 func TestNewAbs(t *testing.T) {
+	t.Parallel()
+
 	testCases := []struct {
 		name string
 
@@ -48,12 +62,14 @@ func TestNewAbs(t *testing.T) {
 		wantErr  error
 	}{
 		{"good", "/etc", MustAbs("/etc"), nil},
-		{"not absolute", "etc", nil, &AbsoluteError{"etc"}},
-		{"zero", "", nil, &AbsoluteError{""}},
+		{"not absolute", "etc", nil, AbsoluteError("etc")},
+		{"zero", "", nil, AbsoluteError("")},
 	}
 
 	for _, tc := range testCases {
 		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
 			got, err := NewAbs(tc.pathname)
 			if !reflect.DeepEqual(got, tc.want) {
 				t.Errorf("NewAbs: %#v, want %#v", got, tc.want)
@@ -65,10 +81,12 @@ func TestNewAbs(t *testing.T) {
 	}
 
 	t.Run("must", func(t *testing.T) {
-		defer func() {
-			wantPanic := `path "etc" is not absolute`
+		t.Parallel()
 
-			if r := recover(); r != wantPanic {
+		defer func() {
+			wantPanic := AbsoluteError("etc")
+
+			if r := recover(); !reflect.DeepEqual(r, wantPanic) {
 				t.Errorf("MustAbs: panic = %v; want %v", r, wantPanic)
 			}
 		}()
@@ -79,13 +97,17 @@ func TestNewAbs(t *testing.T) {
 
 func TestAbsoluteString(t *testing.T) {
 	t.Run("passthrough", func(t *testing.T) {
+		t.Parallel()
+
 		pathname := "/etc"
-		if got := (&Absolute{pathname}).String(); got != pathname {
+		if got := unsafeAbs(pathname).String(); got != pathname {
 			t.Errorf("String: %q, want %q", got, pathname)
 		}
 	})
 
 	t.Run("zero", func(t *testing.T) {
+		t.Parallel()
+
 		defer func() {
 			wantPanic := "attempted use of zero Absolute"
 
@@ -99,6 +121,8 @@ func TestAbsoluteString(t *testing.T) {
 }
 
 func TestAbsoluteIs(t *testing.T) {
+	t.Parallel()
+
 	testCases := []struct {
 		name string
 		a, v *Absolute
@@ -114,6 +138,8 @@ func TestAbsoluteIs(t *testing.T) {
 	}
 	for _, tc := range testCases {
 		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
 			if got := tc.a.Is(tc.v); got != tc.want {
 				t.Errorf("Is: %v, want %v", got, tc.want)
 			}
@@ -123,10 +149,12 @@ func TestAbsoluteIs(t *testing.T) {
 
 type sCheck struct {
 	Pathname *Absolute `json:"val"`
-	Magic    int       `json:"magic"`
+	Magic    uint64    `json:"magic"`
 }
 
 func TestCodecAbsolute(t *testing.T) {
+	t.Parallel()
+
 	testCases := []struct {
 		name string
 		a    *Absolute
@@ -142,32 +170,37 @@ func TestCodecAbsolute(t *testing.T) {
 
 		{"good", MustAbs("/etc"),
 			nil,
-			"\t\x7f\x05\x01\x02\xff\x82\x00\x00\x00\b\xff\x80\x00\x04/etc",
-			",\xff\x83\x03\x01\x01\x06sCheck\x01\xff\x84\x00\x01\x02\x01\bPathname\x01\xff\x80\x00\x01\x05Magic\x01\x04\x00\x00\x00\t\x7f\x05\x01\x02\xff\x82\x00\x00\x00\x10\xff\x84\x01\x04/etc\x01\xfb\x01\x81\xda\x00\x00\x00",
+			"\t\x7f\x06\x01\x02\xff\x82\x00\x00\x00\b\xff\x80\x00\x04/etc",
+			",\xff\x83\x03\x01\x01\x06sCheck\x01\xff\x84\x00\x01\x02\x01\bPathname\x01\xff\x80\x00\x01\x05Magic\x01\x06\x00\x00\x00\t\x7f\x06\x01\x02\xff\x82\x00\x00\x00\x0f\xff\x84\x01\x04/etc\x01\xfc\xc0\xed\x00\x00\x00",
 
 			`"/etc"`, `{"val":"/etc","magic":3236757504}`},
 		{"not absolute", nil,
-			&AbsoluteError{"etc"},
-			"\t\x7f\x05\x01\x02\xff\x82\x00\x00\x00\a\xff\x80\x00\x03etc",
-			",\xff\x83\x03\x01\x01\x06sCheck\x01\xff\x84\x00\x01\x02\x01\bPathname\x01\xff\x80\x00\x01\x05Magic\x01\x04\x00\x00\x00\t\x7f\x05\x01\x02\xff\x82\x00\x00\x00\x0f\xff\x84\x01\x03etc\x01\xfb\x01\x81\xda\x00\x00\x00",
+			AbsoluteError("etc"),
+			"\t\x7f\x06\x01\x02\xff\x82\x00\x00\x00\a\xff\x80\x00\x03etc",
+			",\xff\x83\x03\x01\x01\x06sCheck\x01\xff\x84\x00\x01\x02\x01\bPathname\x01\xff\x80\x00\x01\x05Magic\x01\x06\x00\x00\x00\t\x7f\x06\x01\x02\xff\x82\x00\x00\x00\x0f\xff\x84\x01\x03etc\x01\xfb\x01\x81\xda\x00\x00\x00",
 
 			`"etc"`, `{"val":"etc","magic":3236757504}`},
 		{"zero", nil,
 			new(AbsoluteError),
-			"\t\x7f\x05\x01\x02\xff\x82\x00\x00\x00\x04\xff\x80\x00\x00",
-			",\xff\x83\x03\x01\x01\x06sCheck\x01\xff\x84\x00\x01\x02\x01\bPathname\x01\xff\x80\x00\x01\x05Magic\x01\x04\x00\x00\x00\t\x7f\x05\x01\x02\xff\x82\x00\x00\x00\f\xff\x84\x01\x00\x01\xfb\x01\x81\xda\x00\x00\x00",
+			"\t\x7f\x06\x01\x02\xff\x82\x00\x00\x00\x04\xff\x80\x00\x00",
+			",\xff\x83\x03\x01\x01\x06sCheck\x01\xff\x84\x00\x01\x02\x01\bPathname\x01\xff\x80\x00\x01\x05Magic\x01\x06\x00\x00\x00\t\x7f\x06\x01\x02\xff\x82\x00\x00\x00\f\xff\x84\x01\x00\x01\xfb\x01\x81\xda\x00\x00\x00",
 			`""`, `{"val":"","magic":3236757504}`},
 	}
 
 	for _, tc := range testCases {
 		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
 			t.Run("gob", func(t *testing.T) {
 				if tc.gob == "\x00" && tc.sGob == "\x00" {
 					// these values mark the current test to skip gob
 					return
 				}
+				t.Parallel()
 
 				t.Run("encode", func(t *testing.T) {
+					t.Parallel()
+
 					// encode is unchecked
 					if errors.Is(tc.wantErr, syscall.EINVAL) {
 						return
@@ -204,6 +237,8 @@ func TestCodecAbsolute(t *testing.T) {
 				})
 
 				t.Run("decode", func(t *testing.T) {
+					t.Parallel()
+
 					{
 						var gotA *Absolute
 						err := gob.NewDecoder(strings.NewReader(tc.gob)).Decode(&gotA)
@@ -238,7 +273,11 @@ func TestCodecAbsolute(t *testing.T) {
 			})
 
 			t.Run("json", func(t *testing.T) {
+				t.Parallel()
+
 				t.Run("marshal", func(t *testing.T) {
+					t.Parallel()
+
 					// marshal is unchecked
 					if errors.Is(tc.wantErr, syscall.EINVAL) {
 						return
@@ -273,6 +312,8 @@ func TestCodecAbsolute(t *testing.T) {
 				})
 
 				t.Run("unmarshal", func(t *testing.T) {
+					t.Parallel()
+
 					{
 						var gotA *Absolute
 						err := json.Unmarshal([]byte(tc.json), &gotA)
@@ -306,17 +347,14 @@ func TestCodecAbsolute(t *testing.T) {
 			})
 		})
 	}
-
-	t.Run("json passthrough", func(t *testing.T) {
-		wantErr := "invalid character ':' looking for beginning of value"
-		if err := new(Absolute).UnmarshalJSON([]byte(":3")); err == nil || err.Error() != wantErr {
-			t.Errorf("UnmarshalJSON: error = %v, want %s", err, wantErr)
-		}
-	})
 }
 
 func TestAbsoluteWrap(t *testing.T) {
+	t.Parallel()
+
 	t.Run("join", func(t *testing.T) {
+		t.Parallel()
+
 		want := "/etc/nix/nix.conf"
 		if got := MustAbs("/etc").Append("nix", "nix.conf"); got.String() != want {
 			t.Errorf("Append: %q, want %q", got, want)
@@ -324,6 +362,8 @@ func TestAbsoluteWrap(t *testing.T) {
 	})
 
 	t.Run("dir", func(t *testing.T) {
+		t.Parallel()
+
 		want := "/"
 		if got := MustAbs("/etc").Dir(); got.String() != want {
 			t.Errorf("Dir: %q, want %q", got, want)
@@ -331,6 +371,8 @@ func TestAbsoluteWrap(t *testing.T) {
 	})
 
 	t.Run("sort", func(t *testing.T) {
+		t.Parallel()
+
 		want := []*Absolute{MustAbs("/etc"), MustAbs("/proc"), MustAbs("/sys")}
 		got := []*Absolute{MustAbs("/proc"), MustAbs("/sys"), MustAbs("/etc")}
 		SortAbs(got)
@@ -340,6 +382,8 @@ func TestAbsoluteWrap(t *testing.T) {
 	})
 
 	t.Run("compact", func(t *testing.T) {
+		t.Parallel()
+
 		want := []*Absolute{MustAbs("/etc"), MustAbs("/proc"), MustAbs("/sys")}
 		if got := CompactAbs([]*Absolute{MustAbs("/etc"), MustAbs("/proc"), MustAbs("/proc"), MustAbs("/sys")}); !reflect.DeepEqual(got, want) {
 			t.Errorf("CompactAbs: %#v, want %#v", got, want)

check/overlay.go

@@ -0,0 +1,38 @@
+package check
+
+import "strings"
+
+const (
+	// SpecialOverlayEscape is the escape string for overlay mount options.
+	//
+	// Deprecated: This is no longer used and will be removed in 0.5.
+	SpecialOverlayEscape = `\`
+	// SpecialOverlayOption is the separator string between overlay mount options.
+	//
+	// Deprecated: This is no longer used and will be removed in 0.5.
+	SpecialOverlayOption = ","
+	// SpecialOverlayPath is the separator string between overlay paths.
+	//
+	// Deprecated: This is no longer used and will be removed in 0.5.
+	SpecialOverlayPath = ":"
+)
+
+// EscapeOverlayDataSegment escapes a string for formatting into the data
+// argument of an overlay mount system call.
+//
+// Deprecated: This is no longer used and will be removed in 0.5.
+func EscapeOverlayDataSegment(s string) string {
+	if s == "" {
+		return ""
+	}
+
+	if f := strings.SplitN(s, "\x00", 2); len(f) > 0 {
+		s = f[0]
+	}
+
+	return strings.NewReplacer(
+		SpecialOverlayEscape, SpecialOverlayEscape+SpecialOverlayEscape,
+		SpecialOverlayOption, SpecialOverlayEscape+SpecialOverlayOption,
+		SpecialOverlayPath, SpecialOverlayEscape+SpecialOverlayPath,
+	).Replace(s)
+}

check/overlay_test.go

@@ -0,0 +1,31 @@
+package check_test
+
+import (
+	"testing"
+
+	"hakurei.app/check"
+)
+
+func TestEscapeOverlayDataSegment(t *testing.T) {
+	t.Parallel()
+
+	testCases := []struct {
+		name string
+		s    string
+		want string
+	}{
+		{"zero", "", ""},
+		{"multi", `\\\:,:,\\\`, `\\\\\\\:\,\:\,\\\\\\`},
+		{"bwrap", `/path :,\`, `/path \:\,\\`},
+	}
+
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			if got := check.EscapeOverlayDataSegment(tc.s); got != tc.want {
+				t.Errorf("escapeOverlayDataSegment: %s, want %s", got, tc.want)
+			}
+		})
+	}
+}

cmd/app/app.go

@@ -0,0 +1,264 @@
+package main
+
+import (
+	"bufio"
+	"fmt"
+	"io"
+	"strconv"
+	"strings"
+
+	"hakurei.app/check"
+	"hakurei.app/fhs"
+	"hakurei.app/hst"
+)
+
+// parsePair parses a NUL-delimited quoted paths pair.
+func parsePair(s string) (source, target *check.Absolute, err error) {
+	var p string
+	if p, err = strconv.Unquote(s); err != nil {
+		return
+	}
+	_source, _target, ok := strings.Cut(p, "\x00")
+	if source, err = check.NewAbs(_source); err != nil {
+		return
+	}
+	if !ok {
+		return
+	}
+	target, err = check.NewAbs(_target)
+	return
+}
+
+// parse decodes a high-level configuration stream and returns its
+// corresponding [hst.Config].
+func parse(id string, base *check.Absolute, r io.Reader) (*hst.Config, error) {
+	shell := fhs.AbsRoot.Append("bin", "zsh")
+	home := hst.AbsPrivateTmp.Append("home")
+
+	c := hst.Config{
+		ID:          id,
+		Enablements: new(hst.Enablements),
+
+		SessionBus: &hst.BusConfig{
+			Own: []string{
+				id + ".*",
+				"org.mpris.MediaPlayer2." + id + ".*",
+			},
+			Filter: true,
+		},
+		SystemBus: &hst.BusConfig{Filter: true},
+
+		Container: &hst.ContainerConfig{
+			Env: make(map[string]string),
+			Filesystem: []hst.FilesystemConfigJSON{
+				{FilesystemConfig: &hst.FSOverlay{
+					Target: fhs.AbsRoot,
+					Lower: []*check.Absolute{
+						base.Append("template", "initial"),
+					},
+					Upper: base.Append("template", "upper"),
+				}},
+				{FilesystemConfig: &hst.FSBind{
+					Target: home,
+					Source: base.Append("state", id),
+					Write:  true,
+					Ensure: true,
+				}},
+
+				{FilesystemConfig: &hst.FSEphemeral{
+					Target: fhs.AbsVar.Append("tmp"),
+					Write:  true,
+					Perm:   01777,
+				}},
+
+				{FilesystemConfig: &hst.FSBind{Source: fhs.AbsSys.Append("block")}},
+				{FilesystemConfig: &hst.FSBind{Source: fhs.AbsSys.Append("bus")}},
+				{FilesystemConfig: &hst.FSBind{Source: fhs.AbsSys.Append("class")}},
+				{FilesystemConfig: &hst.FSBind{Source: fhs.AbsSys.Append("dev")}},
+				{FilesystemConfig: &hst.FSBind{Source: fhs.AbsSys.Append("devices")}},
+			},
+
+			Username: "chronos",
+			Shell:    shell,
+			Home:     home,
+			Path:     shell,
+			Args:     []string{"zsh", "-c"},
+
+			Flags: hst.FCoverRun,
+		},
+	}
+
+	s := bufio.NewScanner(r)
+	scanOnce := func() error {
+		if s.Scan() {
+			return nil
+		}
+		if err := s.Err(); err != nil {
+			return err
+		}
+		return io.ErrUnexpectedEOF
+	}
+
+	if err := scanOnce(); err != nil {
+		return nil, err
+	}
+	if v, err := strconv.Atoi(s.Text()); err != nil {
+		return nil, err
+	} else {
+		c.Identity = v
+	}
+
+	if err := scanOnce(); err != nil {
+		return nil, err
+	}
+	c.Container.Args = append(c.Container.Args, s.Text())
+
+	var flagGPU, flagSystemBus bool
+	flags := map[string]*bool{
+		"gpu":        &flagGPU,
+		"system_bus": &flagSystemBus,
+	}
+
+	for s.Scan() {
+		key, value, ok := strings.Cut(s.Text(), " ")
+		if key != "" && key[0] == ';' {
+			continue
+		}
+
+		if !ok {
+			if key == "" {
+				continue
+			}
+
+			var p *bool
+			if p, ok = flags[key]; ok {
+				*p = true
+				continue
+			}
+
+			switch key {
+			case "wayland":
+				*c.Enablements |= hst.EWayland
+			case "x11":
+				*c.Enablements |= hst.EX11
+			case "dbus":
+				*c.Enablements |= hst.EDBus
+			case "pipewire":
+				*c.Enablements |= hst.EPipeWire
+
+			case "multiarch":
+				c.Container.Flags |= hst.FMultiarch
+			case "devel":
+				c.Container.Flags |= hst.FDevel
+			case "userns":
+				c.Container.Flags |= hst.FUserns
+			case "net":
+				c.Container.Flags |= hst.FHostNet
+			case "abstract":
+				c.Container.Flags |= hst.FHostAbstract
+			case "tty":
+				c.Container.Flags |= hst.FTty
+			case "mapuid":
+				c.Container.Flags |= hst.FMapRealUID
+			case "device":
+				c.Container.Flags |= hst.FDevice
+
+			case "share_runtime":
+				c.Container.Flags |= hst.FShareRuntime
+			case "share_tmpdir":
+				c.Container.Flags |= hst.FShareTmpdir
+
+			default:
+				return nil, fmt.Errorf("invalid flag %q", key)
+			}
+
+			continue
+		}
+
+		switch key {
+		case "group":
+			c.Groups = append(c.Groups, value)
+			continue
+
+		case "env":
+			if key, value, ok = strings.Cut(value, "="); !ok {
+				return nil, fmt.Errorf("invalid environment %q", key)
+			}
+			c.Container.Env[key] = value
+			continue
+
+		case "ro":
+			source, target, err := parsePair(value)
+			if err != nil {
+				return nil, err
+			}
+			c.Container.Filesystem = append(c.Container.Filesystem,
+				hst.FilesystemConfigJSON{FilesystemConfig: &hst.FSBind{
+					Target: target,
+					Source: source,
+				}},
+			)
+			continue
+
+		case "rw":
+			source, target, err := parsePair(value)
+			if err != nil {
+				return nil, err
+			}
+			c.Container.Filesystem = append(c.Container.Filesystem,
+				hst.FilesystemConfigJSON{FilesystemConfig: &hst.FSBind{
+					Target: target,
+					Source: source,
+					Write:  true,
+				}},
+			)
+			continue
+
+		case "own":
+			c.SessionBus.Own = append(c.SessionBus.Own, value)
+			continue
+		case "own_system":
+			c.SystemBus.Own = append(c.SystemBus.Own, value)
+			continue
+
+		case "talk":
+			c.SessionBus.Talk = append(c.SessionBus.Talk, value)
+			continue
+		case "talk_system":
+			c.SystemBus.Talk = append(c.SystemBus.Talk, value)
+			continue
+
+		default:
+			return nil, fmt.Errorf("invalid key %q", key)
+		}
+	}
+	if err := s.Err(); err != nil {
+		return nil, err
+	}
+
+	if flagGPU {
+		c.Container.Filesystem = append(c.Container.Filesystem, []hst.FilesystemConfigJSON{
+			{FilesystemConfig: &hst.FSBind{
+				Source:   fhs.AbsDev.Append("dri"),
+				Device:   true,
+				Optional: true,
+			}},
+		}...)
+	}
+
+	if !flagSystemBus {
+		c.SystemBus = nil
+	}
+
+	if c.Container.Flags&hst.FShareTmpdir == 0 {
+		c.Container.Filesystem = append(c.Container.Filesystem,
+			hst.FilesystemConfigJSON{FilesystemConfig: &hst.FSEphemeral{
+				Target: fhs.AbsTmp,
+				Write:  true,
+				Perm:   01777,
+			}},
+		)
+	}
+
+	return &c, nil
+}

cmd/app/app_test.go

@@ -0,0 +1,152 @@
+package main
+
+import (
+	"reflect"
+	"strings"
+	"testing"
+
+	"hakurei.app/check"
+	"hakurei.app/fhs"
+	"hakurei.app/hst"
+)
+
+func TestParse(t *testing.T) {
+	t.Parallel()
+
+	base := fhs.AbsProc.Append("nonexistent")
+	testCases := []struct {
+		name string
+		data string
+		want *hst.Config
+		err  error
+	}{
+		{"com.discordapp.Discord", `8
+exec Discord --ozone-platform-hint=wayland
+
+gpu
+wayland
+dbus
+system_bus
+pipewire
+userns
+net
+mapuid
+
+share_runtime
+share_tmpdir
+
+group media_rw
+env ELECTRON_TRASH=gio
+rw "/sdcard"
+; remove before reusing
+ro "/bin\x00/.hakurei/bin"
+
+talk org.kde.StatusNotifierWatcher
+talk com.canonical.AppMenu.Registrar
+talk com.canonical.indicator.application
+talk com.canonical.Unity
+`, &hst.Config{
+			Identity:    8,
+			ID:          "com.discordapp.Discord",
+			Enablements: new(hst.EWayland | hst.EDBus | hst.EPipeWire),
+			Groups:      []string{"media_rw"},
+
+			SessionBus: &hst.BusConfig{
+				Talk: []string{
+					"org.kde.StatusNotifierWatcher",
+					"com.canonical.AppMenu.Registrar",
+					"com.canonical.indicator.application",
+					"com.canonical.Unity",
+				},
+				Own: []string{
+					"com.discordapp.Discord.*",
+					"org.mpris.MediaPlayer2.com.discordapp.Discord.*",
+				},
+				Filter: true,
+			},
+			SystemBus: &hst.BusConfig{Filter: true},
+
+			Container: &hst.ContainerConfig{
+				Env: map[string]string{
+					"ELECTRON_TRASH": "gio",
+				},
+				Filesystem: []hst.FilesystemConfigJSON{
+					{FilesystemConfig: &hst.FSOverlay{
+						Target: fhs.AbsRoot,
+						Lower: []*check.Absolute{
+							base.Append("template", "initial"),
+						},
+						Upper: base.Append("template", "upper"),
+					}},
+					{FilesystemConfig: &hst.FSBind{
+						Target: hst.AbsPrivateTmp.Append("home"),
+						Source: base.Append("state", "com.discordapp.Discord"),
+						Write:  true,
+						Ensure: true,
+					}},
+
+					{FilesystemConfig: &hst.FSEphemeral{
+						Target: fhs.AbsVar.Append("tmp"),
+						Write:  true,
+						Perm:   01777,
+					}},
+
+					{FilesystemConfig: &hst.FSBind{Source: fhs.AbsSys.Append("block")}},
+					{FilesystemConfig: &hst.FSBind{Source: fhs.AbsSys.Append("bus")}},
+					{FilesystemConfig: &hst.FSBind{Source: fhs.AbsSys.Append("class")}},
+					{FilesystemConfig: &hst.FSBind{Source: fhs.AbsSys.Append("dev")}},
+					{FilesystemConfig: &hst.FSBind{Source: fhs.AbsSys.Append("devices")}},
+
+					{FilesystemConfig: &hst.FSBind{
+						Source: check.MustAbs("/sdcard"),
+						Write:  true,
+					}},
+					{FilesystemConfig: &hst.FSBind{
+						Target: check.MustAbs("/.hakurei/bin"),
+						Source: check.MustAbs("/bin"),
+					}},
+
+					{FilesystemConfig: &hst.FSBind{
+						Source:   fhs.AbsDev.Append("dri"),
+						Device:   true,
+						Optional: true,
+					}},
+				},
+
+				Username: "chronos",
+				Shell:    fhs.AbsRoot.Append("bin", "zsh"),
+				Home:     hst.AbsPrivateTmp.Append("home"),
+				Path:     fhs.AbsRoot.Append("bin", "zsh"),
+				Args: []string{
+					"zsh", "-c",
+					"exec Discord --ozone-platform-hint=wayland",
+				},
+
+				Flags: hst.FCoverRun | hst.FUserns | hst.FHostNet | hst.FMapRealUID |
+					hst.FShareRuntime | hst.FShareTmpdir,
+			},
+		}, nil},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			got, err := parse(
+				tc.name,
+				base,
+				strings.NewReader(tc.data),
+			)
+
+			if !reflect.DeepEqual(err, tc.err) {
+				t.Errorf("parse: error = %v, want %v", err, tc.err)
+			}
+			if err != nil {
+				return
+			}
+
+			if !reflect.DeepEqual(got, tc.want) {
+				t.Errorf("parse: %#v, want %#v", got, tc.want)
+			}
+		})
+	}
+}

cmd/app/main.go

@@ -0,0 +1,170 @@
+// The app program is a proof-of-concept frontend for cmd/hakurei.
+//
+// This program is not covered by the compatibility promise. The command line
+// interface and configuration syntax may change at any time.
+package main
+
+import (
+	"context"
+	"errors"
+	"log"
+	"os"
+	"os/exec"
+	"os/signal"
+	"path/filepath"
+	"syscall"
+
+	"hakurei.app/check"
+	"hakurei.app/command"
+	"hakurei.app/fhs"
+	"hakurei.app/hst"
+	"hakurei.app/message"
+)
+
+func main() {
+	log.SetFlags(0)
+	log.SetPrefix("app: ")
+	msg := message.New(log.Default())
+
+	ctx, stop := signal.NotifyContext(context.Background(),
+		syscall.SIGINT, syscall.SIGTERM, syscall.SIGHUP)
+	defer stop()
+
+	var (
+		flagVerbose bool
+		flagBase    string
+
+		base, template, initial, upper, work *check.Absolute
+	)
+	c := command.New(os.Stderr, log.Printf, "app", func([]string) (err error) {
+		msg.SwapVerbose(flagVerbose)
+		flagBase = os.ExpandEnv(flagBase)
+		if flagBase == "" {
+			flagBase = "state"
+		}
+		if flagBase, err = filepath.Abs(flagBase); err != nil {
+			return
+		} else if base, err = check.NewAbs(flagBase); err != nil {
+			return
+		}
+
+		template = base.Append("template")
+		initial = template.Append("initial")
+		upper = template.Append("upper")
+		work = template.Append("work")
+		return
+	}).Flag(
+		&flagVerbose,
+		"v", command.BoolFlag(false),
+		"Increase log verbosity",
+	).Flag(
+		&flagBase,
+		"d", command.StringFlag("$HAKUREI_APP_PATH"),
+		"Configuration and state directory",
+	)
+
+	{
+		var (
+			flagShell string
+			flagHome  string
+		)
+		c.NewCommand(
+			"enter", "Enter mutable state template",
+			func([]string) error {
+				config := hst.Config{
+					ID: "app.hakurei.mutable",
+					Container: &hst.ContainerConfig{
+						Hostname: "mutable",
+						Filesystem: []hst.FilesystemConfigJSON{
+							{FilesystemConfig: &hst.FSOverlay{
+								Target: fhs.AbsRoot,
+								Lower:  []*check.Absolute{initial},
+								Upper:  upper,
+								Work:   work,
+							}},
+							{FilesystemConfig: &hst.FSEphemeral{
+								Target: fhs.AbsTmp,
+								Write:  true,
+								Perm:   0755,
+							}},
+						},
+						Username: "chronos",
+						Flags: hst.FMultiarch |
+							hst.FDevel |
+							hst.FUserns |
+							hst.FHostNet |
+							hst.FTty,
+					},
+				}
+
+				if a, err := check.NewAbs(flagShell); err != nil {
+					return err
+				} else {
+					config.Container.Shell = a
+					config.Container.Path = a
+					config.Container.Args = []string{
+						"-" + filepath.Base(flagShell),
+					}
+				}
+
+				if a, err := check.NewAbs(flagHome); err != nil {
+					return err
+				} else {
+					config.Container.Home = a
+				}
+
+				return run(ctx, msg, &config)
+			},
+		).Flag(
+			&flagShell,
+			"shell", command.StringFlag("/bin/zsh"),
+			"Shell program within container",
+		).Flag(
+			&flagHome,
+			"home", command.StringFlag("/home/chronos"),
+			"Home directory within container",
+		)
+	}
+
+	c.NewCommand(
+		"run", "Start the named application",
+		func(args []string) error {
+			if len(args) != 1 {
+				return errors.New("run requires 1 argument")
+			}
+
+			var config *hst.Config
+			f, err := os.Open(base.Append("app", args[0]).String())
+			if err != nil {
+				return err
+			}
+			config, err = parse(args[0], base, f)
+			if closeErr := f.Close(); err == nil {
+				err = closeErr
+			}
+			if err != nil {
+				return err
+			}
+
+			return run(ctx, msg, config)
+		},
+	)
+
+	c.MustParse(os.Args[1:], func(err error) {
+		if e, ok := errors.AsType[*exec.ExitError](err); ok && e != nil {
+			os.Exit(e.ExitCode())
+		}
+
+		if w, ok := err.(interface{ Unwrap() []error }); !ok {
+			log.Fatal(err)
+		} else {
+			errs := w.Unwrap()
+			for i, e := range errs {
+				if i == len(errs)-1 {
+					log.Fatal(e)
+				}
+				log.Println(e)
+			}
+		}
+	})
+}

cmd/app/run.go

@@ -0,0 +1,51 @@
+package main
+
+import (
+	"context"
+	"encoding/json"
+	"os"
+	"os/exec"
+	"syscall"
+
+	"hakurei.app/hst"
+	"hakurei.app/message"
+)
+
+// run starts a container via cmd/hakurei and returns after it terminates.
+func run(ctx context.Context, msg message.Msg, config *hst.Config) error {
+	c, cancel := context.WithCancel(ctx)
+	defer cancel()
+
+	cmd := exec.CommandContext(c, "hakurei")
+	cmd.Stdin, cmd.Stdout, cmd.Stderr = os.Stdin, os.Stdout, os.Stderr
+	cmd.Cancel = func() error {
+		return cmd.Process.Signal(syscall.SIGINT)
+	}
+	if msg.IsVerbose() {
+		cmd.Args = append(cmd.Args, "-v")
+	}
+	cmd.Args = append(cmd.Args, "run", "3")
+
+	r, w, err := os.Pipe()
+	if err != nil {
+		return err
+	}
+	cmd.ExtraFiles = append(cmd.ExtraFiles, r)
+
+	if err = cmd.Start(); err != nil {
+		_, _ = r.Close(), w.Close()
+		return err
+	}
+
+	if err = r.Close(); err != nil {
+		_ = w.Close()
+		return err
+	} else if err = json.NewEncoder(w).Encode(&config); err != nil {
+		_ = w.Close()
+		return err
+	} else if err = w.Close(); err != nil {
+		return err
+	}
+
+	return cmd.Wait()
+}

cmd/dist/VERSION

@@ -0,0 +1 @@
+v0.4.4

cmd/dist/comp/_hakurei

@@ -1,21 +1,24 @@
 #compdef hakurei
 
-_hakurei_app() {
+_hakurei_run() {
   __hakurei_files
   return $?
 }
 
-_hakurei_run() {
+_hakurei_exec() {
   _arguments \
     '--id[Reverse-DNS style Application identifier, leave empty to inherit instance identifier]:id' \
     '-a[Application identity]: :_numbers' \
     '-g[Groups inherited by all container processes]: :_groups' \
     '-d[Container home directory]: :_files -/' \
     '-u[Passwd user name within sandbox]: :_users' \
+    '--private-runtime[Do not share XDG_RUNTIME_DIR between containers under the same identity]' \
+    '--private-tmpdir[Do not share TMPDIR between containers under the same identity]' \
     '--wayland[Enable connection to Wayland via security-context-v1]' \
     '-X[Enable direct connection to X11]' \
     '--dbus[Enable proxied connection to D-Bus]' \
-    '--pulse[Enable direct connection to PulseAudio]' \
+    '--pipewire[Enable connection to PipeWire via SecurityContext]' \
+    '--pulse[Enable PulseAudio compatibility daemon]' \
     '--dbus-config[Path to session bus proxy config file]: :_files -g "*.json"' \
     '--dbus-system[Path to system bus proxy config file]: :_files -g "*.json"' \
     '--mpris[Allow owning MPRIS D-Bus path]' \
@@ -54,9 +57,9 @@ __hakurei_instances() {
 {
   local -a _hakurei_cmds
   _hakurei_cmds=(
-    "app:Load app from configuration file"
-    "run:Configure and start a permissive default sandbox"
-    "show:Show live or local app configuration"
+    "run:Load and start container from configuration file"
+    "exec:Configure and start a permissive container"
+    "show:Show live or local instance configuration"
     "ps:List active instances"
     "version:Display version information"
     "license:Show full license text"

cmd/dist/dist.sh

@@ -0,0 +1,10 @@
+#!/bin/sh -e
+
+TOOLCHAIN_VERSION="$(go version)"
+cd "$(dirname -- "$0")/../.."
+echo "Building cmd/dist using ${TOOLCHAIN_VERSION}."
+FLAGS=''
+if test -n "$VERBOSE"; then
+    FLAGS="$FLAGS -v"
+fi
+go run $FLAGS --tags=dist ./cmd/dist

cmd/dist/main.go

@@ -0,0 +1,254 @@
+//go:build dist
+
+package main
+
+import (
+	"archive/tar"
+	"compress/gzip"
+	"context"
+	"crypto/sha512"
+	_ "embed"
+	"encoding/hex"
+	"fmt"
+	"io"
+	"io/fs"
+	"log"
+	"os"
+	"os/exec"
+	"os/signal"
+	"path/filepath"
+	"runtime"
+	"strings"
+)
+
+//go:generate sh -c "git describe --tags > VERSION"
+//go:embed VERSION
+var version string
+
+// getenv looks up an environment variable, and returns fallback if it is unset.
+func getenv(key, fallback string) string {
+	if v, ok := os.LookupEnv(key); ok {
+		return v
+	}
+	return fallback
+}
+
+// mustRun runs a command with the current process's environment and panics
+// on error or non-zero exit code.
+func mustRun(ctx context.Context, name string, arg ...string) {
+	cmd := exec.CommandContext(ctx, name, arg...)
+	cmd.Stdin, cmd.Stdout, cmd.Stderr = os.Stdin, os.Stdout, os.Stderr
+	if err := cmd.Run(); err != nil {
+		panic(err)
+	}
+}
+
+//go:embed comp/_hakurei
+var comp []byte
+
+func main() {
+	log.SetFlags(0)
+	log.SetPrefix("")
+
+	verbose := os.Getenv("VERBOSE") != ""
+	runTests := os.Getenv("HAKUREI_DIST_MAKE") == ""
+	version = getenv("HAKUREI_VERSION", strings.TrimSpace(version))
+	prefix := getenv("PREFIX", "/usr")
+	destdir := getenv("DESTDIR", "dist")
+
+	if verbose {
+		log.Println()
+	}
+
+	if err := os.MkdirAll(destdir, 0755); err != nil {
+		log.Fatal(err)
+	}
+	s, err := os.MkdirTemp(destdir, ".dist.*")
+	if err != nil {
+		log.Fatal(err)
+	}
+	defer func() {
+		var code int
+
+		if err = os.RemoveAll(s); err != nil {
+			code = 1
+			log.Println(err)
+		}
+
+		if r := recover(); r != nil {
+			code = 1
+			log.Println(r)
+		}
+
+		os.Exit(code)
+	}()
+
+	ctx, cancel := signal.NotifyContext(context.Background(), os.Interrupt)
+	defer cancel()
+
+	verboseFlag := "-v"
+	if !verbose {
+		verboseFlag = "-buildvcs=false"
+	}
+
+	log.Printf("Building hakurei for %s/%s.", runtime.GOOS, runtime.GOARCH)
+	mustRun(ctx, "go", "generate", "./...")
+	mustRun(
+		ctx, "go", "build",
+		"-trimpath",
+		verboseFlag, "-o", s,
+		"-ldflags=-s -w "+
+			"-buildid= -linkmode external -extldflags=-static "+
+			"-X hakurei.app/internal/info.buildVersion="+version+" "+
+			"-X hakurei.app/internal/info.hakureiPath="+prefix+"/bin/hakurei "+
+			"-X hakurei.app/internal/info.hsuPath="+prefix+"/bin/hsu "+
+			"-X main.hakureiPath="+prefix+"/bin/hakurei",
+		"./...",
+	)
+	log.Println()
+
+	if runTests {
+		log.Println("##### Testing Hakurei.")
+		mustRun(
+			ctx, "go", "test",
+			"-ldflags=-buildid= -linkmode external -extldflags=-static",
+			"./...",
+		)
+		log.Println()
+	}
+
+	log.Println("##### Creating distribution.")
+	const suffix = ".tar.gz"
+	distName := "hakurei-" + version + "-" + runtime.GOARCH
+	var f *os.File
+	if f, err = os.OpenFile(
+		filepath.Join(s, distName+suffix),
+		os.O_CREATE|os.O_EXCL|os.O_WRONLY,
+		0644,
+	); err != nil {
+		panic(err)
+	}
+	defer func() {
+		if f == nil {
+			return
+		}
+		if err = f.Close(); err != nil {
+			log.Println(err)
+		}
+	}()
+
+	h := sha512.New()
+	gw, _ := gzip.NewWriterLevel(io.MultiWriter(f, h), gzip.BestCompression)
+	tw := tar.NewWriter(gw)
+
+	mustWriteHeader := func(name string, size int64, mode os.FileMode) {
+		header := tar.Header{
+			Name:  filepath.Join(distName, name),
+			Size:  size,
+			Mode:  int64(mode),
+			Uname: "root",
+			Gname: "root",
+		}
+
+		if mode&os.ModeDir != 0 {
+			header.Typeflag = tar.TypeDir
+			fmt.Printf("%s %s\n", mode, name)
+		} else {
+			header.Typeflag = tar.TypeReg
+			fmt.Printf("%s %s (%d bytes)\n", mode, name, size)
+		}
+
+		if err = tw.WriteHeader(&header); err != nil {
+			panic(err)
+		}
+	}
+
+	mustWriteFile := func(name string, data []byte, mode os.FileMode) {
+		mustWriteHeader(name, int64(len(data)), mode)
+		if mode&os.ModeDir != 0 {
+			return
+		}
+		if _, err = tw.Write(data); err != nil {
+			panic(err)
+		}
+	}
+
+	mustWriteFromPath := func(dst, src string, mode os.FileMode) {
+		var r *os.File
+		if r, err = os.Open(src); err != nil {
+			panic(err)
+		}
+
+		var fi os.FileInfo
+		if fi, err = r.Stat(); err != nil {
+			_ = r.Close()
+			panic(err)
+		}
+
+		if mode == 0 {
+			mode = fi.Mode()
+		}
+
+		mustWriteHeader(dst, fi.Size(), mode)
+		if _, err = io.Copy(tw, r); err != nil {
+			_ = r.Close()
+			panic(err)
+		} else if err = r.Close(); err != nil {
+			panic(err)
+		}
+	}
+
+	mustWriteFile(".", nil, fs.ModeDir|0755)
+	mustWriteFile("comp/", nil, os.ModeDir|0755)
+	mustWriteFile("comp/_hakurei", comp, 0644)
+	mustWriteFile("install.sh", []byte(`#!/bin/sh -e
+cd "$(dirname -- "$0")" || exit 1
+
+install -vDm0755 "bin/hakurei" "${DESTDIR}`+prefix+`/bin/hakurei"
+install -vDm0755 "bin/sharefs" "${DESTDIR}`+prefix+`/bin/sharefs"
+
+install -vDm4511 "bin/hsu" "${DESTDIR}`+prefix+`/bin/hsu"
+if [ ! -f "${DESTDIR}/etc/hsurc" ]; then
+    install -vDm0400 "hsurc.default" "${DESTDIR}/etc/hsurc"
+fi
+
+install -vDm0644 "comp/_hakurei" "${DESTDIR}`+prefix+`/share/zsh/site-functions/_hakurei"
+`), 0755)
+
+	mustWriteFromPath("README.md", "README.md", 0)
+	mustWriteFile("hsurc.default", []byte("1000 0"), 0400)
+	mustWriteFromPath("bin/hsu", filepath.Join(s, "hsu"), 04511)
+	for _, name := range []string{
+		"hakurei",
+		"sharefs",
+	} {
+		mustWriteFromPath(
+			filepath.Join("bin", name),
+			filepath.Join(s, name),
+			0,
+		)
+	}
+
+	if err = tw.Close(); err != nil {
+		panic(err)
+	} else if err = gw.Close(); err != nil {
+		panic(err)
+	} else if err = f.Close(); err != nil {
+		panic(err)
+	}
+	f = nil
+
+	if err = os.WriteFile(
+		filepath.Join(destdir, distName+suffix+".sha512"),
+		append(hex.AppendEncode(nil, h.Sum(nil)), "  "+distName+suffix+"\n"...),
+		0644,
+	); err != nil {
+		panic(err)
+	}
+	if err = os.Rename(
+		filepath.Join(s, distName+suffix),
+		filepath.Join(destdir, distName+suffix),
+	); err != nil {
+		panic(err)
+	}
+}

cmd/earlyinit/main.go

@@ -0,0 +1,264 @@
+// The earlyinit is part of the Rosa OS initramfs and serves as the system init.
+//
+// This program is an internal detail of Rosa OS and is not usable on its own.
+// It is not covered by the compatibility promise.
+package main
+
+import (
+	"context"
+	"crypto/rand"
+	"log"
+	"os"
+	"os/signal"
+	"runtime"
+	"runtime/pprof"
+	"slices"
+	"strings"
+	. "syscall"
+
+	"hakurei.app/internal/kobject"
+	"hakurei.app/internal/report"
+	"hakurei.app/internal/uevent"
+	"hakurei.app/message"
+)
+
+var r report.Reporter
+
+func init() {
+	log.SetFlags(0)
+	log.SetPrefix("earlyinit: ")
+	r.SetOutput(log.Default())
+
+	// this handles SIGQUIT to provide useful debugging information without
+	// terminating, and prevents the runtime from throwing on the must family
+	// of early error reporting functions, DO NOT REMOVE
+	c := make(chan os.Signal, 1)
+	signal.Notify(c, SIGQUIT)
+	go func() {
+		for {
+			<-c
+			if p := pprof.Lookup("goroutine"); p == nil {
+				log.Println("initial built-in goroutine profile does not exist")
+			} else if err := p.WriteTo(os.Stderr, 2); err != nil {
+				log.Println(err)
+			}
+		}
+	}()
+}
+
+// fatal calls [log.Println] with v and blocks forever. Must be called from
+// main. Must not be used after error reporting is set up.
+func fatal(v ...any) {
+	log.Println(v...)
+	log.Println("unable to continue, please reboot and resolve the problem manually")
+	select {}
+}
+
+// must calls fatal with err if it is non-nil.
+func must(err error) {
+	if err != nil {
+		log.Println(err)
+		select {}
+	}
+}
+
+// mustSyscall is like must, but with an additional action name.
+func mustSyscall(action string, err error) {
+	if err != nil {
+		fatal("cannot "+action+":", err)
+		select {}
+	}
+}
+
+// must1 is like must, but with an additional passed through value.
+func must1[T any](v T, err error) T {
+	must(err)
+	return v
+}
+
+const (
+	// optionSystem specifies devpath of the system device.
+	optionSystem = "system"
+
+	// flagVerbose increases output verbosity.
+	flagVerbose = "verbose"
+	// flagStrict sets [report.DStrict] on r.
+	flagStrict = "strict"
+	// flagNoRecover sets [report.DNoRecover] on r.
+	flagNoRecover = "no_recover"
+)
+
+func main() {
+	runtime.LockOSThread()
+
+	var (
+		option map[string]string
+		flags  []string
+	)
+	if len(os.Args) > 1 {
+		option = make(map[string]string)
+		for _, s := range os.Args[1:] {
+			key, value, ok := strings.Cut(s, "=")
+			if !ok {
+				flags = append(flags, s)
+				continue
+			}
+			option[key] = value
+		}
+	}
+
+	{
+		var flag uint64
+		if slices.Contains(flags, flagStrict) {
+			flag |= report.DStrict
+		}
+		if slices.Contains(flags, flagNoRecover) {
+			flag |= report.DNoRecover
+		}
+		log.Printf("reporting flags %x", flag)
+		r.SetFlags(flag)
+	}
+
+	msg := message.New(log.Default())
+	msg.SwapVerbose(slices.Contains(flags, flagVerbose))
+
+	mustSyscall("mount devtmpfs", Mount(
+		"devtmpfs",
+		"/dev/",
+		"devtmpfs",
+		MS_NOSUID|MS_NOEXEC,
+		"",
+	))
+	must(os.Mkdir("/dev/pts/", 0))
+	mustSyscall("mount devpts", Mount(
+		"devpts",
+		"/dev/pts/",
+		"devpts",
+		MS_NOSUID|MS_NOEXEC,
+		"mode=620,ptmxmode=666",
+	))
+	must(os.Mkdir("/dev/shm/", 0))
+	mustSyscall("mount shm", Mount(
+		"shm",
+		"/dev/shm/",
+		"tmpfs",
+		MS_NOSUID|MS_NODEV,
+		"",
+	))
+
+	// The kernel might be unable to set up the console. When that happens,
+	// printk is called with "Warning: unable to open an initial console."
+	// and the init runs with no files. The checkfds runtime function
+	// populates 0-2 by opening /dev/null for them.
+	//
+	// This check replaces 1 and 2 with /dev/kmsg to improve the chance
+	// of output being visible to the user.
+	if fi, err := os.Stdout.Stat(); err == nil {
+		if stat, ok := fi.Sys().(*Stat_t); ok {
+			if stat.Rdev == 0x103 {
+				var fd int
+				if fd, err = Open(
+					"/dev/kmsg",
+					O_WRONLY|O_CLOEXEC,
+					0,
+				); err != nil {
+					log.Fatalf("cannot open kmsg: %v", err)
+				}
+
+				if err = Dup3(fd, Stdout, 0); err != nil {
+					log.Fatalf("cannot open stdout: %v", err)
+				}
+				if err = Dup3(fd, Stderr, 0); err != nil {
+					log.Fatalf("cannot open stderr: %v", err)
+				}
+
+				if err = Close(fd); err != nil {
+					log.Printf("cannot close kmsg: %v", err)
+				}
+			}
+		}
+	}
+
+	// staying in rootfs, these are no longer used
+	must(os.Remove("/root"))
+	must(os.Remove("/init"))
+
+	must(os.Mkdir("/proc", 0))
+	mustSyscall("mount proc", Mount(
+		"proc",
+		"/proc",
+		"proc",
+		MS_NOSUID|MS_NOEXEC|MS_NODEV,
+		"hidepid=1",
+	))
+
+	must(os.Mkdir("/sys", 0))
+	mustSyscall("mount sysfs", Mount(
+		"sysfs",
+		"/sys",
+		"sysfs",
+		0,
+		"",
+	))
+
+	conn := must1(uevent.Dial(-128 * 1024 * 1024))
+	events := make(chan *uevent.Message, 1<<10)
+	var uuid uevent.UUID
+	must1(rand.Read(uuid[:]))
+	ctx, cancel := context.WithCancel(context.Background())
+
+	go consume(ctx, msg, &r, conn, uuid, events)
+	s := kobject.New(uuid, func(o *kobject.Object, env map[string]string) {
+		p := make([]string, 0, len(env))
+		for k, v := range env {
+			p = append(p, k+"="+v)
+		}
+		slices.Sort(p)
+		log.Printf("change %s: %s", o.DevPath, strings.Join(p, ", "))
+	}, func(err error) {
+		severity := report.Inconsistent
+		if e, ok := err.(kobject.EventError); ok && e.Kind == kobject.EBadTarget {
+			severity = report.Trivial
+		}
+		r.Dispatch(
+			severity,
+			"processed inconsistent uevent",
+			err,
+		)
+	})
+	go func() {
+		s.Consume(ctx, events)
+
+		log.Println("closing NETLINK_KOBJECT_UEVENT socket")
+		cancel()
+		if err := conn.Close(); err != nil {
+			log.Fatal(err) // not reached
+		}
+	}()
+
+	must(os.Mkdir("/system", 0))
+	if devpath := option[optionSystem]; devpath == "" {
+		fatal("system must be nonempty")
+	} else {
+		log.Printf("waiting for devpath pattern %q", devpath)
+		mustMountSystem(ctx, s, devpath)
+	}
+
+	// after top level has been set up
+	mustSyscall("remount root", Mount(
+		"",
+		"/",
+		"",
+		MS_REMOUNT|MS_BIND|
+			MS_RDONLY|MS_NODEV|MS_NOSUID|MS_NOEXEC,
+		"",
+	))
+
+	must(os.WriteFile(
+		"/sys/module/firmware_class/parameters/path",
+		[]byte("/system/lib/firmware"),
+		0,
+	))
+	go dispatchModprobe(ctx, s)
+
+}

cmd/earlyinit/modprobe.go

@@ -0,0 +1,73 @@
+package main
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"log"
+	"os/exec"
+	"strings"
+
+	"hakurei.app/internal/kobject"
+	"hakurei.app/internal/report"
+	"hakurei.app/internal/uevent"
+)
+
+// ModprobeError describes an unsuccessful modprobe invocation.
+type ModprobeError struct {
+	ModAlias string `json:"modalias"`
+	Stdout   string `json:"stdout"`
+	Stderr   string `json:"stderr"`
+	ExitCode int    `json:"exit_code"`
+}
+
+var _ report.RepresentableError = ModprobeError{}
+
+func (ModprobeError) Representable() {}
+func (e ModprobeError) Error() string {
+	return fmt.Sprintf(
+		"modprobe exit status %d: %s",
+		e.ExitCode, strings.TrimSpace(e.Stderr),
+	)
+}
+
+// dispatchModprobe invokes modprobe for [uevent.KOBJ_ADD] events raising new
+// MODALIAS strings.
+func dispatchModprobe(
+	ctx context.Context,
+	s *kobject.State,
+) {
+	aliases := make(chan string, 1<<8)
+	go func() {
+		defer close(aliases)
+		s.Range(ctx, func(o *kobject.Object, act uevent.KobjectAction) bool {
+			if act == uevent.KOBJ_ADD && o.Driver == "" && o.ModAlias != "" {
+				aliases <- o.ModAlias
+			}
+			return true
+		})
+	}()
+
+	for alias := range aliases {
+		stdout, err := exec.Command("/system/sbin/modprobe", alias).Output()
+		if err == nil {
+			if len(stdout) > 0 {
+				log.Println(string(stdout))
+			}
+			continue
+		}
+
+		exitError, ok := errors.AsType[*exec.ExitError](err)
+		if !ok || exitError == nil {
+			r.Dispatch(report.Degraded, "invoke modprobe", err)
+			continue
+		}
+
+		r.Dispatch(report.Trivial, "load device driver", ModprobeError{
+			ModAlias: alias,
+			Stdout:   string(stdout),
+			Stderr:   string(exitError.Stderr),
+			ExitCode: exitError.ExitCode(),
+		})
+	}
+}

cmd/earlyinit/mount.go

@@ -0,0 +1,71 @@
+package main
+
+import (
+	"context"
+	"errors"
+	"os"
+	"path/filepath"
+	"strconv"
+	"syscall"
+	"time"
+
+	"hakurei.app/check"
+	"hakurei.app/fhs"
+	"hakurei.app/internal/kobject"
+	"hakurei.app/internal/uevent"
+)
+
+// mustMountSystem waits for and mounts a system device matching pattern.
+func mustMountSystem(
+	ctx context.Context,
+	s *kobject.State,
+	pattern string,
+) {
+	c, stop := context.WithTimeout(ctx, 30*time.Second)
+	defer stop()
+
+	for {
+		var matchErr error
+		var systemPath *check.Absolute
+		s.Range(c, func(o *kobject.Object, act uevent.KobjectAction) bool {
+			if (act != uevent.KOBJ_ADD && act != uevent.KOBJ_CHANGE) ||
+				o.Subsystem != "block" ||
+				o.Env["DEVTYPE"] != "disk" {
+				return true
+			}
+
+			if ok, err := filepath.Match(pattern, o.DevPath); err != nil {
+				matchErr = err
+				return false
+			} else if !ok {
+				return true
+			}
+
+			name, ok := o.Env["DEVNAME"]
+			if !ok {
+				return true
+			}
+			systemPath = fhs.AbsDev.Append(name)
+			return false
+		})
+		if c.Err() != nil {
+			fatal("devpath", strconv.Quote(pattern), "never appeared")
+		}
+		if matchErr != nil {
+			fatal("cannot match system devpath:", matchErr)
+		}
+		err := syscall.Mount(
+			systemPath.String(),
+			"/system/",
+			"squashfs",
+			0,
+			"threads=multi",
+		)
+		if err == nil {
+			break
+		}
+		if !errors.Is(err, os.ErrNotExist) {
+			fatal("cannot mount system:", err)
+		}
+	}
+}

cmd/earlyinit/uevent.go

@@ -0,0 +1,104 @@
+package main
+
+import (
+	"context"
+	"time"
+
+	"hakurei.app/fhs"
+	"hakurei.app/internal/report"
+	"hakurei.app/internal/uevent"
+	"hakurei.app/message"
+)
+
+// newRejectColdboot returns a function to be called on every subsequent pending
+// coldboot, and returns whether coldboot should proceed. Rejection is sticky.
+func newRejectColdboot() func() bool {
+	// one coldboot per five minutes, two consecutive coldboot
+	const (
+		coldbootInterval = 5 * time.Minute
+		coldbootBurst    = 2
+	)
+
+	done := make(chan struct{})
+	s := make(chan struct{}, coldbootBurst)
+	s <- struct{}{} // for early fault before reporting is ready
+	go func() {
+		t := time.NewTicker(coldbootInterval)
+		for {
+			select {
+			case <-done:
+				return
+
+			case <-t.C:
+				select {
+				case s <- struct{}{}:
+				default:
+				}
+			}
+		}
+	}()
+
+	return func() bool {
+		select {
+		case <-s:
+			return true
+
+		case <-done:
+			return false
+
+		default:
+			close(done)
+			return false
+		}
+	}
+}
+
+// consume continuously consumes events from conn with retries.
+func consume(
+	ctx context.Context,
+	msg message.Msg,
+	r *report.Reporter,
+	conn *uevent.Conn,
+	uuid uevent.UUID,
+	events chan<- *uevent.Message,
+) {
+	defer close(events)
+
+	nextColdboot := newRejectColdboot()
+	coldboot := true
+retry:
+	if dispatchErr := conn.Consume(ctx, fhs.Sys, &uuid, events, coldboot, func(path string) {
+		msg.Verbose("coldboot visited", path)
+	}, func(err error) bool {
+		if _, ok := err.(uevent.NeedsColdboot); ok && !nextColdboot() {
+			r.Dispatch(
+				report.Degraded,
+				"rejecting coldboot loop",
+				err,
+			)
+			return false
+		}
+		r.Dispatch(
+			report.Inconsistent,
+			"consumed invalid message",
+			err,
+		)
+		return true
+	}, nil); dispatchErr != nil {
+		if _, ok := dispatchErr.(uevent.Recoverable); !ok {
+			r.Dispatch(
+				report.Fatal,
+				"discontinuing uevent processing due to nonrecoverable error",
+				dispatchErr,
+			)
+			return
+		}
+
+		if _, ok := dispatchErr.(uevent.NeedsColdboot); ok {
+			// coldboot loop rejected by handler
+			coldboot = false
+		}
+
+		goto retry
+	}
+}

cmd/earlyinit/uevent_test.go

@@ -0,0 +1,35 @@
+package main
+
+import (
+	"testing"
+	"testing/synctest"
+	"time"
+)
+
+func TestRejectColdboot(t *testing.T) {
+	t.Parallel()
+
+	synctest.Test(t, func(t *testing.T) {
+		nextColdboot := newRejectColdboot()
+		want := func(want bool) {
+			if got := nextColdboot(); got != want {
+				t.Fatalf("nextColdboot: %v, want %v", got, want)
+			}
+		}
+
+		synctest.Wait()
+		want(true)
+		time.Sleep(time.Hour)
+		synctest.Wait()
+		want(true)
+		want(true)
+		time.Sleep(5 * time.Minute)
+		synctest.Wait()
+		want(true)
+		want(false)
+		time.Sleep(time.Hour)
+		synctest.Wait()
+		want(false)
+		want(false)
+	})
+}

cmd/hakurei/command.go

@@ -2,77 +2,117 @@ package main
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"io"
 	"log"
 	"os"
-	"os/signal"
+	"os/exec"
 	"os/user"
 	"strconv"
 	"sync"
-	"syscall"
 	"time"
 
+	"hakurei.app/check"
 	"hakurei.app/command"
-	"hakurei.app/container"
+	"hakurei.app/ext"
+	"hakurei.app/fhs"
 	"hakurei.app/hst"
-	"hakurei.app/internal"
-	"hakurei.app/internal/app"
-	"hakurei.app/internal/app/state"
-	"hakurei.app/internal/hlog"
-	"hakurei.app/system"
-	"hakurei.app/system/dbus"
+	"hakurei.app/internal/dbus"
+	"hakurei.app/internal/env"
+	"hakurei.app/internal/info"
+	"hakurei.app/internal/outcome"
+	"hakurei.app/message"
 )
 
-func buildCommand(out io.Writer) command.Command {
+// optionalErrorUnwrap calls [errors.Unwrap] and returns the resulting value
+// if it is not nil, or the original value if it is.
+func optionalErrorUnwrap(err error) error {
+	if underlyingErr := errors.Unwrap(err); underlyingErr != nil {
+		return underlyingErr
+	}
+	return err
+}
+
+var errSuccess = errors.New("success")
+
+func buildCommand(ctx context.Context, msg message.Msg, early *earlyHardeningErrs, out io.Writer) command.Command {
 	var (
-		flagVerbose bool
-		flagJSON    bool
+		flagVerbose  bool
+		flagInsecure bool
+		flagJSON     bool
 	)
-	c := command.New(out, log.Printf, "hakurei", func([]string) error { internal.InstallOutput(flagVerbose); return nil }).
-		Flag(&flagVerbose, "v", command.BoolFlag(false), "Increase log verbosity").
-		Flag(&flagJSON, "json", command.BoolFlag(false), "Serialise output in JSON when applicable")
+	c := command.New(out, log.Printf, "hakurei", func([]string) error {
+		msg.SwapVerbose(flagVerbose)
 
-	c.Command("shim", command.UsageInternal, func([]string) error { app.ShimMain(); return errSuccess })
-
-	c.Command("app", "Load app from configuration file", func(args []string) error {
-		if len(args) < 1 {
-			log.Fatal("app requires at least 1 argument")
+		if early.yamaLSM != nil {
+			msg.Verbosef("cannot enable ptrace protection via Yama LSM: %v", early.yamaLSM)
+			// not fatal
 		}
 
-		// config extraArgs...
-		config := tryPath(args[0])
-		config.Args = append(config.Args, args[1:]...)
+		if early.dumpable != nil {
+			log.Printf("cannot set SUID_DUMP_DISABLE: %s", early.dumpable)
+			// not fatal
+		}
 
-		runApp(config)
-		panic("unreachable")
-	})
+		return nil
+	}).
+		Flag(&flagVerbose, "v", command.BoolFlag(false), "Increase log verbosity").
+		Flag(&flagInsecure, "insecure", command.BoolFlag(false), "Allow use of insecure compatibility options").
+		Flag(&flagJSON, "json", command.BoolFlag(false), "Serialise output in JSON when applicable")
+
+	c.Command("shim", command.UsageInternal, func([]string) error { outcome.Shim(msg); return errSuccess })
 
 	{
 		var (
-			dbusConfigSession string
-			dbusConfigSystem  string
-			mpris             bool
-			dbusVerbose       bool
-
-			fid      string
-			aid      int
-			groups   command.RepeatableFlag
-			homeDir  string
-			userName string
-
-			wayland, x11, dBus, pulse bool
+			flagIdentifierFile int
 		)
-
-		c.NewCommand("run", "Configure and start a permissive default sandbox", func(args []string) error {
-			// initialise config from flags
-			config := &hst.Config{
-				ID:   fid,
-				Args: args,
+		c.NewCommand("run", "Load and start container from configuration file", func(args []string) error {
+			if len(args) < 1 {
+				log.Fatal("run requires at least 1 argument")
 			}
 
-			if aid < 0 || aid > 9999 {
-				log.Fatalf("aid %d out of range", aid)
+			config := tryPath(msg, args[0])
+			if config != nil && config.Container != nil {
+				config.Container.Args = append(config.Container.Args, args[1:]...)
+			}
+
+			var flags int
+			if flagInsecure {
+				flags |= hst.VAllowInsecure
+			}
+
+			outcome.Main(ctx, msg, config, flags, flagIdentifierFile)
+			panic("unreachable")
+		}).
+			Flag(&flagIdentifierFile, "identifier-fd", command.IntFlag(-1),
+				"Write identifier of current instance to fd after successful startup")
+	}
+
+	{
+		var (
+			flagDBusConfigSession string
+			flagDBusConfigSystem  string
+			flagDBusMpris         bool
+			flagDBusVerbose       bool
+
+			flagID       string
+			flagIdentity int
+			flagGroups   command.RepeatableFlag
+			flagHomeDir  string
+			flagUserName string
+
+			flagSchedPolicy   string
+			flagSchedPriority int
+
+			flagPrivateRuntime, flagPrivateTmpdir bool
+
+			flagWayland, flagX11, flagDBus, flagPipeWire, flagPulse bool
+		)
+
+		c.NewCommand("exec", "Configure and start a permissive container", func(args []string) error {
+			if flagIdentity < hst.IdentityStart || flagIdentity > hst.IdentityEnd {
+				log.Fatalf("identity %d out of range", flagIdentity)
 			}
 
 			// resolve home/username from os when flag is unset
@@ -80,22 +120,15 @@ func buildCommand(out io.Writer) command.Command {
 				passwd     *user.User
 				passwdOnce sync.Once
 				passwdFunc = func() {
-					var us string
-					if uid, err := std.Uid(aid); err != nil {
-						hlog.PrintBaseError(err, "cannot obtain uid from setuid wrapper:")
-						os.Exit(1)
-					} else {
-						us = strconv.Itoa(uid)
-					}
-
+					us := strconv.Itoa(hst.ToUser(new(outcome.Hsu).MustID(msg), flagIdentity))
 					if u, err := user.LookupId(us); err != nil {
-						hlog.Verbosef("cannot look up uid %s", us)
+						msg.Verbosef("cannot look up uid %s", us)
 						passwd = &user.User{
 							Uid:      us,
 							Gid:      us,
 							Username: "chronos",
 							Name:     "Hakurei Permissive Default",
-							HomeDir:  container.FHSVarEmpty,
+							HomeDir:  fhs.VarEmpty,
 						}
 					} else {
 						passwd = u
@@ -103,164 +136,256 @@ func buildCommand(out io.Writer) command.Command {
 				}
 			)
 
-			if homeDir == "os" {
+			// paths are identical, resolve inner shell and program path
+			shell := fhs.AbsRoot.Append("bin", "sh")
+			if a, err := check.NewAbs(os.Getenv("SHELL")); err == nil {
+				shell = a
+			}
+			progPath := shell
+			if len(args) > 0 {
+				if p, err := exec.LookPath(args[0]); err != nil {
+					log.Fatal(optionalErrorUnwrap(err))
+					return err
+				} else if progPath, err = check.NewAbs(p); err != nil {
+					log.Fatal(err)
+					return err
+				}
+			}
+
+			var et hst.Enablements
+			if flagWayland {
+				et |= hst.EWayland
+			}
+			if flagX11 {
+				et |= hst.EX11
+			}
+			if flagDBus {
+				et |= hst.EDBus
+			}
+			if flagPipeWire || flagPulse {
+				et |= hst.EPipeWire
+			}
+
+			config := hst.Config{
+				ID:          flagID,
+				Identity:    flagIdentity,
+				Groups:      flagGroups,
+				Enablements: &et,
+
+				Container: &hst.ContainerConfig{
+					Filesystem: []hst.FilesystemConfigJSON{
+						// autoroot, includes the home directory
+						{FilesystemConfig: &hst.FSBind{
+							Target:  fhs.AbsRoot,
+							Source:  fhs.AbsRoot,
+							Write:   true,
+							Special: true,
+						}},
+					},
+
+					Username: flagUserName,
+					Shell:    shell,
+
+					Path: progPath,
+					Args: args,
+
+					Flags: hst.FUserns | hst.FHostNet | hst.FHostAbstract | hst.FTty,
+				},
+			}
+
+			if err := config.SchedPolicy.UnmarshalText(
+				[]byte(flagSchedPolicy),
+			); err != nil {
+				log.Fatal(err)
+			}
+			config.SchedPriority = ext.Int(flagSchedPriority)
+
+			// bind GPU stuff
+			if et&(hst.EX11|hst.EWayland) != 0 {
+				config.Container.Filesystem = append(config.Container.Filesystem, hst.FilesystemConfigJSON{FilesystemConfig: &hst.FSBind{
+					Source:   fhs.AbsDev.Append("dri"),
+					Device:   true,
+					Optional: true,
+				}})
+			}
+
+			config.Container.Filesystem = append(config.Container.Filesystem,
+				// opportunistically bind kvm
+				hst.FilesystemConfigJSON{FilesystemConfig: &hst.FSBind{
+					Source:   fhs.AbsDev.Append("kvm"),
+					Device:   true,
+					Optional: true,
+				}},
+
+				// do autoetc last
+				hst.FilesystemConfigJSON{FilesystemConfig: &hst.FSBind{
+					Target:  fhs.AbsEtc,
+					Source:  fhs.AbsEtc,
+					Special: true,
+				}},
+			)
+
+			if config.Container.Username == "chronos" {
 				passwdOnce.Do(passwdFunc)
-				homeDir = passwd.HomeDir
+				config.Container.Username = passwd.Username
 			}
 
-			if userName == "chronos" {
-				passwdOnce.Do(passwdFunc)
-				userName = passwd.Username
+			{
+				homeDir := flagHomeDir
+				if homeDir == "os" {
+					passwdOnce.Do(passwdFunc)
+					homeDir = passwd.HomeDir
+				}
+				if a, err := check.NewAbs(homeDir); err != nil {
+					log.Fatal(err)
+					return err
+				} else {
+					config.Container.Home = a
+				}
 			}
 
-			config.Identity = aid
-			config.Groups = groups
-			config.Username = userName
-
-			if a, err := container.NewAbs(homeDir); err != nil {
-				log.Fatal(err.Error())
-				return err
-			} else {
-				config.Home = a
+			if !flagPrivateRuntime {
+				config.Container.Flags |= hst.FShareRuntime
 			}
-
-			var e system.Enablement
-			if wayland {
-				e |= system.EWayland
+			if !flagPrivateTmpdir {
+				config.Container.Flags |= hst.FShareTmpdir
 			}
-			if x11 {
-				e |= system.EX11
-			}
-			if dBus {
-				e |= system.EDBus
-			}
-			if pulse {
-				e |= system.EPulse
-			}
-			config.Enablements = hst.NewEnablements(e)
 
 			// parse D-Bus config file from flags if applicable
-			if dBus {
-				if dbusConfigSession == "builtin" {
-					config.SessionBus = dbus.NewConfig(fid, true, mpris)
+			if flagDBus {
+				if flagDBusConfigSession == "builtin" {
+					config.SessionBus = dbus.NewConfig(flagID, true, flagDBusMpris)
 				} else {
-					if conf, err := dbus.NewConfigFromFile(dbusConfigSession); err != nil {
-						log.Fatalf("cannot load session bus proxy config from %q: %s", dbusConfigSession, err)
+					if f, err := os.Open(flagDBusConfigSession); err != nil {
+						log.Fatal(err)
 					} else {
-						config.SessionBus = conf
+						decodeJSON(log.Fatal, "load session bus proxy config", f, &config.SessionBus)
+						if err = f.Close(); err != nil {
+							log.Fatal(err)
+						}
 					}
 				}
 
 				// system bus proxy is optional
-				if dbusConfigSystem != "nil" {
-					if conf, err := dbus.NewConfigFromFile(dbusConfigSystem); err != nil {
-						log.Fatalf("cannot load system bus proxy config from %q: %s", dbusConfigSystem, err)
+				if flagDBusConfigSystem != "nil" {
+					if f, err := os.Open(flagDBusConfigSystem); err != nil {
+						log.Fatal(err)
 					} else {
-						config.SystemBus = conf
+						decodeJSON(log.Fatal, "load system bus proxy config", f, &config.SystemBus)
+						if err = f.Close(); err != nil {
+							log.Fatal(err)
+						}
 					}
 				}
 
 				// override log from configuration
-				if dbusVerbose {
-					config.SessionBus.Log = true
-					config.SystemBus.Log = true
+				if flagDBusVerbose {
+					if config.SessionBus != nil {
+						config.SessionBus.Log = true
+					}
+					if config.SystemBus != nil {
+						config.SystemBus.Log = true
+					}
 				}
 			}
 
-			// invoke app
-			runApp(config)
+			outcome.Main(ctx, msg, &config, 0, -1)
 			panic("unreachable")
 		}).
-			Flag(&dbusConfigSession, "dbus-config", command.StringFlag("builtin"),
+			Flag(&flagDBusConfigSession, "dbus-config", command.StringFlag("builtin"),
 				"Path to session bus proxy config file, or \"builtin\" for defaults").
-			Flag(&dbusConfigSystem, "dbus-system", command.StringFlag("nil"),
+			Flag(&flagDBusConfigSystem, "dbus-system", command.StringFlag("nil"),
 				"Path to system bus proxy config file, or \"nil\" to disable").
-			Flag(&mpris, "mpris", command.BoolFlag(false),
+			Flag(&flagDBusMpris, "mpris", command.BoolFlag(false),
 				"Allow owning MPRIS D-Bus path, has no effect if custom config is available").
-			Flag(&dbusVerbose, "dbus-log", command.BoolFlag(false),
+			Flag(&flagDBusVerbose, "dbus-log", command.BoolFlag(false),
 				"Force buffered logging in the D-Bus proxy").
-			Flag(&fid, "id", command.StringFlag(""),
+			Flag(&flagID, "id", command.StringFlag(""),
 				"Reverse-DNS style Application identifier, leave empty to inherit instance identifier").
-			Flag(&aid, "a", command.IntFlag(0),
+			Flag(&flagIdentity, "a", command.IntFlag(0),
 				"Application identity").
-			Flag(nil, "g", &groups,
+			Flag(nil, "g", &flagGroups,
 				"Groups inherited by all container processes").
-			Flag(&homeDir, "d", command.StringFlag("os"),
+			Flag(&flagHomeDir, "d", command.StringFlag("os"),
 				"Container home directory").
-			Flag(&userName, "u", command.StringFlag("chronos"),
+			Flag(&flagUserName, "u", command.StringFlag("chronos"),
 				"Passwd user name within sandbox").
-			Flag(&wayland, "wayland", command.BoolFlag(false),
+			Flag(&flagSchedPolicy, "policy", command.StringFlag(""),
+				"Scheduling policy to set for the container").
+			Flag(&flagSchedPriority, "priority", command.IntFlag(0),
+				"Scheduling priority to set for the container").
+			Flag(&flagPrivateRuntime, "private-runtime", command.BoolFlag(false),
+				"Do not share XDG_RUNTIME_DIR between containers under the same identity").
+			Flag(&flagPrivateTmpdir, "private-tmpdir", command.BoolFlag(false),
+				"Do not share TMPDIR between containers under the same identity").
+			Flag(&flagWayland, "wayland", command.BoolFlag(false),
 				"Enable connection to Wayland via security-context-v1").
-			Flag(&x11, "X", command.BoolFlag(false),
+			Flag(&flagX11, "X", command.BoolFlag(false),
 				"Enable direct connection to X11").
-			Flag(&dBus, "dbus", command.BoolFlag(false),
+			Flag(&flagDBus, "dbus", command.BoolFlag(false),
 				"Enable proxied connection to D-Bus").
-			Flag(&pulse, "pulse", command.BoolFlag(false),
-				"Enable direct connection to PulseAudio")
+			Flag(&flagPipeWire, "pipewire", command.BoolFlag(false),
+				"Enable connection to PipeWire via SecurityContext").
+			Flag(&flagPulse, "pulse", command.BoolFlag(false),
+				"Enable PulseAudio compatibility daemon")
 	}
 
-	var showFlagShort bool
-	c.NewCommand("show", "Show live or local app configuration", func(args []string) error {
-		switch len(args) {
-		case 0: // system
-			printShowSystem(os.Stdout, showFlagShort, flagJSON)
+	{
+		var (
+			flagShort   bool
+			flagNoStore bool
+		)
+		c.NewCommand("show", "Show live or local instance configuration", func(args []string) error {
+			switch len(args) {
+			case 0: // system
+				printShowSystem(os.Stdout, flagShort, flagJSON)
 
-		case 1: // instance
-			name := args[0]
-			config, entry := tryShort(name)
-			if config == nil {
-				config = tryPath(name)
+			case 1: // instance
+				name := args[0]
+
+				var (
+					config *hst.Config
+					entry  *hst.State
+				)
+				if !flagNoStore {
+					var sc hst.Paths
+					env.CopyPaths().Copy(&sc, new(outcome.Hsu).MustID(nil))
+					entry = tryIdentifier(msg, name, outcome.NewStore(&sc))
+				}
+
+				if entry == nil {
+					config = tryPath(msg, name)
+				} else {
+					config = entry.Config
+				}
+
+				if !printShowInstance(os.Stdout, time.Now().UTC(), entry, config, flagShort, flagJSON) {
+					os.Exit(1)
+				}
+
+			default:
+				log.Fatal("show requires 1 argument")
 			}
-			printShowInstance(os.Stdout, time.Now().UTC(), entry, config, showFlagShort, flagJSON)
+			return errSuccess
+		}).
+			Flag(&flagShort, "short", command.BoolFlag(false), "Omit filesystem information").
+			Flag(&flagNoStore, "no-store", command.BoolFlag(false), "Do not attempt to match from active instances")
+	}
 
-		default:
-			log.Fatal("show requires 1 argument")
-		}
-		return errSuccess
-	}).Flag(&showFlagShort, "short", command.BoolFlag(false), "Omit filesystem information")
+	{
+		var flagShort bool
+		c.NewCommand("ps", "List active instances", func(args []string) error {
+			var sc hst.Paths
+			env.CopyPaths().Copy(&sc, new(outcome.Hsu).MustID(nil))
+			printPs(msg, os.Stdout, time.Now().UTC(), outcome.NewStore(&sc), flagShort, flagJSON)
+			return errSuccess
+		}).Flag(&flagShort, "short", command.BoolFlag(false), "Print instance id")
+	}
 
-	var psFlagShort bool
-	c.NewCommand("ps", "List active instances", func(args []string) error {
-		printPs(os.Stdout, time.Now().UTC(), state.NewMulti(std.Paths().RunDirPath.String()), psFlagShort, flagJSON)
-		return errSuccess
-	}).Flag(&psFlagShort, "short", command.BoolFlag(false), "Print instance id")
-
-	c.Command("version", "Display version information", func(args []string) error {
-		fmt.Println(internal.Version())
-		return errSuccess
-	})
-
-	c.Command("license", "Show full license text", func(args []string) error {
-		fmt.Println(license)
-		return errSuccess
-	})
-
-	c.Command("template", "Produce a config template", func(args []string) error {
-		printJSON(os.Stdout, false, hst.Template())
-		return errSuccess
-	})
-
-	c.Command("help", "Show this help message", func([]string) error {
-		c.PrintHelp()
-		return errSuccess
-	})
+	c.Command("version", "Display version information", func(args []string) error { fmt.Println(info.Version()); return errSuccess })
+	c.Command("license", "Show full license text", func(args []string) error { fmt.Println(license); return errSuccess })
+	c.Command("template", "Produce a config template", func(args []string) error { encodeJSON(log.Fatal, os.Stdout, false, hst.Template()); return errSuccess })
+	c.Command("help", "Show this help message", func([]string) error { c.PrintHelp(); return errSuccess })
 
 	return c
 }
-
-func runApp(config *hst.Config) {
-	ctx, stop := signal.NotifyContext(context.Background(),
-		syscall.SIGINT, syscall.SIGTERM)
-	defer stop() // unreachable
-	a := app.MustNew(ctx, std)
-
-	rs := new(app.RunState)
-	if sa, err := a.Seal(config); err != nil {
-		hlog.PrintBaseError(err, "cannot seal app:")
-		internal.Exit(1)
-	} else {
-		internal.Exit(app.PrintRunStateErr(rs, sa.Run(rs)))
-	}
-
-	*(*int)(nil) = 0 // not reached
-}

cmd/hakurei/command_test.go

@@ -7,9 +7,12 @@ import (
 	"testing"
 
 	"hakurei.app/command"
+	"hakurei.app/message"
 )
 
 func TestHelp(t *testing.T) {
+	t.Parallel()
+
 	testCases := []struct {
 		name string
 		args []string
@@ -17,12 +20,12 @@ func TestHelp(t *testing.T) {
 	}{
 		{
 			"main", []string{}, `
-Usage:	hakurei [-h | --help] [-v] [--json] COMMAND [OPTIONS]
+Usage:	hakurei [-h | --help] [-v] [--insecure] [--json] COMMAND [OPTIONS]
 
 Commands:
-    app         Load app from configuration file
-    run         Configure and start a permissive default sandbox
-    show        Show live or local app configuration
+    run         Load and start container from configuration file
+    exec        Configure and start a permissive container
+    show        Show live or local instance configuration
     ps          List active instances
     version     Display version information
     license     Show full license text
@@ -32,8 +35,8 @@ Commands:
 `,
 		},
 		{
-			"run", []string{"run", "-h"}, `
-Usage:	hakurei run [-h | --help] [--dbus-config <value>] [--dbus-system <value>] [--mpris] [--dbus-log] [--id <value>] [-a <int>] [-g <value>] [-d <value>] [-u <value>] [--wayland] [-X] [--dbus] [--pulse] COMMAND [OPTIONS]
+			"exec", []string{"exec", "-h"}, `
+Usage:	hakurei exec [-h | --help] [--dbus-config <value>] [--dbus-system <value>] [--mpris] [--dbus-log] [--id <value>] [-a <int>] [-g <value>] [-d <value>] [-u <value>] [--policy <value>] [--priority <int>] [--private-runtime] [--private-tmpdir] [--wayland] [-X] [--dbus] [--pipewire] [--pulse] COMMAND [OPTIONS]
 
 Flags:
   -X	Enable direct connection to X11
@@ -55,8 +58,18 @@ Flags:
     	Reverse-DNS style Application identifier, leave empty to inherit instance identifier
   -mpris
     	Allow owning MPRIS D-Bus path, has no effect if custom config is available
+  -pipewire
+    	Enable connection to PipeWire via SecurityContext
+  -policy string
+    	Scheduling policy to set for the container
+  -priority int
+    	Scheduling priority to set for the container
+  -private-runtime
+    	Do not share XDG_RUNTIME_DIR between containers under the same identity
+  -private-tmpdir
+    	Do not share TMPDIR between containers under the same identity
   -pulse
-    	Enable direct connection to PulseAudio
+    	Enable PulseAudio compatibility daemon
   -u string
     	Passwd user name within sandbox (default "chronos")
   -wayland
@@ -67,8 +80,10 @@ Flags:
 	}
 	for _, tc := range testCases {
 		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
 			out := new(bytes.Buffer)
-			c := buildCommand(out)
+			c := buildCommand(t.Context(), message.New(nil), new(earlyHardeningErrs), out)
 			if err := c.Parse(tc.args); !errors.Is(err, command.ErrHelp) && !errors.Is(err, flag.ErrHelp) {
 				t.Errorf("Parse: error = %v; want %v",
 					err, command.ErrHelp)

cmd/hakurei/json.go

@@ -0,0 +1,61 @@
+package main
+
+import (
+	"encoding/json"
+	"errors"
+	"io"
+	"strconv"
+)
+
+// decodeJSON decodes json from r and stores it in v. A non-nil error results in
+// a call to fatal.
+func decodeJSON(fatal func(v ...any), op string, r io.Reader, v any) {
+	err := json.NewDecoder(r).Decode(v)
+	if err == nil {
+		return
+	}
+
+	var (
+		syntaxError        *json.SyntaxError
+		unmarshalTypeError *json.UnmarshalTypeError
+
+		msg string
+	)
+
+	switch {
+	case errors.As(err, &syntaxError) && syntaxError != nil:
+		msg = syntaxError.Error() +
+			" at byte " + strconv.FormatInt(syntaxError.Offset, 10)
+
+	case errors.As(err, &unmarshalTypeError) && unmarshalTypeError != nil:
+		msg = "inappropriate " + unmarshalTypeError.Value +
+			" at byte " + strconv.FormatInt(unmarshalTypeError.Offset, 10)
+
+	default:
+		// InvalidUnmarshalError: incorrect usage, does not need to be handled
+		// io.ErrUnexpectedEOF: no additional error information available
+		msg = err.Error()
+	}
+
+	fatal("cannot " + op + ": " + msg)
+}
+
+// encodeJSON encodes v to output. A non-nil error results in a call to fatal.
+func encodeJSON(fatal func(v ...any), output io.Writer, short bool, v any) {
+	encoder := json.NewEncoder(output)
+	if !short {
+		encoder.SetIndent("", "  ")
+	}
+
+	if err := encoder.Encode(v); err != nil {
+		if e, ok := errors.AsType[*json.MarshalerError](err); ok && e != nil {
+			// this likely indicates an implementation error in hst
+			fatal("cannot encode json for " + e.Type.String() + ": " + e.Err.Error())
+			return
+		}
+
+		// UnsupportedTypeError, UnsupportedValueError: incorrect usage, does
+		// not need to be handled
+		fatal("cannot write json: " + err.Error())
+	}
+}

cmd/hakurei/json_test.go

@@ -0,0 +1,99 @@
+package main
+
+import (
+	"reflect"
+	"strings"
+	"testing"
+
+	"hakurei.app/internal/stub"
+)
+
+func TestDecodeJSON(t *testing.T) {
+	t.Parallel()
+
+	testCases := []struct {
+		name string
+		t    reflect.Type
+		data string
+		want any
+		msg  string
+	}{
+		{"success", reflect.TypeFor[uintptr](), "3735928559\n", uintptr(0xdeadbeef), ""},
+
+		{"syntax", reflect.TypeFor[*int](), "\x00", nil,
+			`cannot load sample: invalid character '\x00' looking for beginning of value at byte 1`},
+		{"type", reflect.TypeFor[uintptr](), "-1", nil,
+			`cannot load sample: inappropriate number -1 at byte 2`},
+		{"default", reflect.TypeFor[*int](), "{", nil,
+			"cannot load sample: unexpected EOF"},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			var (
+				gotP   = reflect.New(tc.t)
+				gotMsg *string
+			)
+			decodeJSON(func(v ...any) {
+				if gotMsg != nil {
+					t.Fatal("fatal called twice")
+				}
+				msg := v[0].(string)
+				gotMsg = &msg
+			}, "load sample", strings.NewReader(tc.data), gotP.Interface())
+			if tc.msg != "" {
+				if gotMsg == nil {
+					t.Errorf("decodeJSON: success, want fatal %q", tc.msg)
+				} else if *gotMsg != tc.msg {
+					t.Errorf("decodeJSON: fatal = %q, want %q", *gotMsg, tc.msg)
+				}
+			} else if gotMsg != nil {
+				t.Errorf("decodeJSON: fatal = %q", *gotMsg)
+			} else if !reflect.DeepEqual(gotP.Elem().Interface(), tc.want) {
+				t.Errorf("decodeJSON: %#v, want %#v", gotP.Elem().Interface(), tc.want)
+			}
+		})
+	}
+}
+
+func TestEncodeJSON(t *testing.T) {
+	t.Parallel()
+
+	testCases := []struct {
+		name string
+		v    any
+		want string
+	}{
+		{"marshaler", errorJSONMarshaler{},
+			`cannot encode json for main.errorJSONMarshaler: unique error 3735928559 injected by the test suite`},
+		{"default", func() {},
+			`cannot write json: json: unsupported type: func()`},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			var called bool
+			encodeJSON(func(v ...any) {
+				if called {
+					t.Fatal("fatal called twice")
+				}
+				called = true
+
+				if v[0].(string) != tc.want {
+					t.Errorf("encodeJSON: fatal = %q, want %q", v[0].(string), tc.want)
+				}
+			}, nil, false, tc.v)
+
+			if !called {
+				t.Errorf("encodeJSON: success, want fatal %q", tc.want)
+			}
+		})
+	}
+}
+
+// errorJSONMarshaler implements json.Marshaler.
+type errorJSONMarshaler struct{}
+
+func (errorJSONMarshaler) MarshalJSON() ([]byte, error) { return nil, stub.UniqueError(0xdeadbeef) }

cmd/hakurei/main.go

@@ -1,53 +1,88 @@
+// Hakurei runs user-specified containers as subordinate users.
+//
+// This program is generally invoked by another, higher level program, which
+// creates container configuration via package [hst] or an implementation of it.
+//
+// The parent may leave files open and specify their file descriptor for various
+// uses. In these cases, standard streams and netpoll files are treated as
+// invalid file descriptors and rejected. All string representations must be in
+// decimal.
+//
+// When specifying a [hst.Config] JSON stream or file to the run subcommand, the
+// argument "-" is equivalent to stdin. Otherwise, file descriptor rules
+// described above applies. Invalid file descriptors are treated as file names
+// in their string representation, with the exception that if a netpoll file
+// descriptor is attempted, the program fails.
+//
+// The flag --identifier-fd can be optionally specified to the run subcommand to
+// receive the identifier of the newly started instance. File descriptor rules
+// described above applies, and the file must be writable. This is sent after
+// its state is made available, so the client must not attempt to poll for it.
+// This uses the internal binary format of [hst.ID].
+//
+// For the show and ps subcommands, the flag --json can be applied to the main
+// hakurei command to serialise output in JSON when applicable. Additionally,
+// the flag --short targeting each subcommand is used to omit some information
+// in both JSON and user-facing output. Only JSON-encoded output is covered
+// under the compatibility promise.
+//
+// A template for [hst.Config] demonstrating all available configuration fields
+// is returned by [hst.Template]. The JSON-encoded equivalent of this can be
+// obtained via the template subcommand. Fields left unpopulated in the template
+// (the direct_* family of fields, which are insecure under any configuration if
+// enabled) are unsupported.
+//
+// For simple (but insecure) testing scenarios, the exec subcommand can be used
+// to generate a simple, permissive configuration in-memory. See its help
+// message for all available options.
 package main
 
-// this works around go:embed '..' limitation
-//go:generate cp ../../LICENSE .
-
 import (
+	"context"
 	_ "embed"
 	"errors"
 	"log"
 	"os"
+	"os/signal"
+	"syscall"
 
 	"hakurei.app/container"
-	"hakurei.app/internal"
-	"hakurei.app/internal/hlog"
-	"hakurei.app/internal/sys"
+	"hakurei.app/ext"
+	"hakurei.app/message"
 )
 
-var (
-	errSuccess = errors.New("success")
+//go:generate cp ../../LICENSE .
+//go:embed LICENSE
+var license string
 
-	//go:embed LICENSE
-	license string
-)
-
-func init() { hlog.Prepare("hakurei") }
-
-var std sys.State = new(sys.Std)
+// earlyHardeningErrs are errors collected while setting up early hardening feature.
+type earlyHardeningErrs struct{ yamaLSM, dumpable error }
 
 func main() {
 	// early init path, skips root check and duplicate PR_SET_DUMPABLE
-	container.TryArgv0(hlog.Output{}, hlog.Prepare, internal.InstallOutput)
+	container.TryArgv0(nil)
 
-	if err := container.SetPtracer(0); err != nil {
-		hlog.Verbosef("cannot enable ptrace protection via Yama LSM: %v", err)
-		// not fatal: this program runs as the privileged user
-	}
+	log.SetFlags(0)
+	log.SetPrefix("hakurei: ")
+	msg := message.New(log.Default())
 
-	if err := container.SetDumpable(container.SUID_DUMP_DISABLE); err != nil {
-		log.Printf("cannot set SUID_DUMP_DISABLE: %s", err)
-		// not fatal: this program runs as the privileged user
+	early := earlyHardeningErrs{
+		yamaLSM:  ext.SetPtracer(0),
+		dumpable: ext.SetDumpable(ext.SUID_DUMP_DISABLE),
 	}
 
 	if os.Geteuid() == 0 {
 		log.Fatal("this program must not run as root")
 	}
 
-	buildCommand(os.Stderr).MustParse(os.Args[1:], func(err error) {
-		hlog.Verbosef("command returned %v", err)
+	ctx, stop := signal.NotifyContext(context.Background(),
+		syscall.SIGINT, syscall.SIGTERM)
+	defer stop() // unreachable
+
+	buildCommand(ctx, msg, &early, os.Stderr).MustParse(os.Args[1:], func(err error) {
+		msg.Verbosef("command returned %v", err)
 		if errors.Is(err, errSuccess) {
-			hlog.BeforeExit()
+			msg.BeforeExit()
 			os.Exit(0)
 		}
 		// this catches faulty command handlers that fail to return before this point

cmd/hakurei/parse.go

@@ -1,7 +1,7 @@
 package main
 
 import (
-	"encoding/json"
+	"encoding/hex"
 	"errors"
 	"io"
 	"log"
@@ -11,66 +11,104 @@ import (
 	"syscall"
 
 	"hakurei.app/hst"
-	"hakurei.app/internal/app/state"
-	"hakurei.app/internal/hlog"
+	"hakurei.app/internal/outcome"
+	"hakurei.app/internal/store"
+	"hakurei.app/message"
 )
 
-func tryPath(name string) (config *hst.Config) {
-	var r io.Reader
+// tryPath attempts to read [hst.Config] from multiple sources.
+//
+// tryPath reads from [os.Stdin] if name has value "-". Otherwise, name is
+// passed to tryFd, and if that returns nil, name is passed to [os.Open].
+func tryPath(msg message.Msg, name string) (config *hst.Config) {
+	var r io.ReadCloser
 	config = new(hst.Config)
 
 	if name != "-" {
-		r = tryFd(name)
+		r = tryFd(msg, name)
 		if r == nil {
-			hlog.Verbose("load configuration from file")
+			msg.Verbose("load configuration from file")
 
 			if f, err := os.Open(name); err != nil {
-				log.Fatalf("cannot access configuration file %q: %s", name, err)
+				log.Fatal(err.Error())
+				return
 			} else {
-				// finalizer closes f
 				r = f
 			}
-		} else {
-			defer func() {
-				if err := r.(io.ReadCloser).Close(); err != nil {
-					log.Printf("cannot close config fd: %v", err)
-				}
-			}()
 		}
 	} else {
 		r = os.Stdin
 	}
 
-	if err := json.NewDecoder(r).Decode(&config); err != nil {
-		log.Fatalf("cannot load configuration: %v", err)
+	decodeJSON(log.Fatal, "load configuration", r, &config)
+	if err := r.Close(); err != nil {
+		log.Fatal(err.Error())
 	}
-
 	return
 }
 
-func tryFd(name string) io.ReadCloser {
+// tryFd returns a [io.ReadCloser] if name represents an integer corresponding
+// to a valid file descriptor.
+func tryFd(msg message.Msg, name string) io.ReadCloser {
 	if v, err := strconv.Atoi(name); err != nil {
 		if !errors.Is(err, strconv.ErrSyntax) {
-			hlog.Verbosef("name cannot be interpreted as int64: %v", err)
+			msg.Verbosef("name cannot be interpreted as int64: %v", err)
 		}
 		return nil
 	} else {
-		hlog.Verbosef("trying config stream from %d", v)
+		if v < 3 { // reject standard streams
+			return nil
+		}
+
+		msg.Verbosef("trying config stream from %d", v)
 		fd := uintptr(v)
-		if _, _, errno := syscall.Syscall(syscall.SYS_FCNTL, fd, syscall.F_GETFD, 0); errno != 0 {
-			if errors.Is(errno, syscall.EBADF) {
+		if _, _, errno := syscall.Syscall(
+			syscall.SYS_FCNTL,
+			fd,
+			syscall.F_GETFD,
+			0,
+		); errno != 0 {
+			if errors.Is(errno, syscall.EBADF) { // reject bad fd
 				return nil
 			}
 			log.Fatalf("cannot get fd %d: %v", fd, errno)
 		}
+
+		if outcome.IsPollDescriptor(fd) { // reject runtime internals
+			log.Fatalf("invalid config stream %d", fd)
+		}
+
 		return os.NewFile(fd, strconv.Itoa(v))
 	}
 }
 
-func tryShort(name string) (config *hst.Config, entry *state.State) {
-	likePrefix := false
-	if len(name) <= 32 {
-		likePrefix = true
+// shortLengthMin is the minimum length a short form identifier can have and
+// still be interpreted as an identifier.
+const shortLengthMin = 1 << 3
+
+// shortIdentifier returns an eight character short representation of [hst.ID]
+// from its random bytes.
+func shortIdentifier(id *hst.ID) string {
+	return shortIdentifierString(id.String())
+}
+
+// shortIdentifierString implements shortIdentifier on an arbitrary string.
+func shortIdentifierString(s string) string {
+	return s[len(hst.ID{}) : len(hst.ID{})+shortLengthMin]
+}
+
+// tryIdentifier attempts to match [hst.State] from a [hex] representation of
+// [hst.ID] or a prefix of its lower half.
+func tryIdentifier(msg message.Msg, name string, s *store.Store) *hst.State {
+	const (
+		likeShort = 1 << iota
+		likeFull
+	)
+
+	var likely uintptr
+	// half the hex representation
+	if len(name) >= shortLengthMin && len(name) <= len(hst.ID{}) {
+		// cannot safely decode here due to unknown alignment
 		for _, c := range name {
 			if c >= '0' && c <= '9' {
 				continue
@@ -78,33 +116,68 @@ func tryShort(name string) (config *hst.Config, entry *state.State) {
 			if c >= 'a' && c <= 'f' {
 				continue
 			}
-			likePrefix = false
-			break
+			return nil
 		}
+		likely |= likeShort
+	} else if len(name) == hex.EncodedLen(len(hst.ID{})) {
+		likely |= likeFull
 	}
 
-	// try to match from state store
-	if likePrefix && len(name) >= 8 {
-		hlog.Verbose("argument looks like prefix")
+	if likely == 0 {
+		return nil
+	}
 
-		s := state.NewMulti(std.Paths().RunDirPath.String())
-		if entries, err := state.Join(s); err != nil {
-			log.Printf("cannot join store: %v", err)
-			// drop to fetch from file
-		} else {
-			for id := range entries {
-				v := id.String()
-				if strings.HasPrefix(v, name) {
-					// match, use config from this state entry
-					entry = entries[id]
-					config = entry.Config
-					break
+	entries, copyError := s.All()
+	defer func() {
+		if err := copyError(); err != nil {
+			msg.GetLogger().Println(getMessage("cannot iterate over store:", err))
+		}
+	}()
+
+	switch {
+	case likely&likeShort != 0:
+		msg.Verbose("argument looks like short identifier")
+		for eh := range entries {
+			if eh.DecodeErr != nil {
+				msg.Verbose(getMessage("skipping instance:", eh.DecodeErr))
+				continue
+			}
+
+			if strings.HasPrefix(eh.ID.String()[len(hst.ID{}):], name) {
+				var entry hst.State
+				if _, err := eh.Load(&entry); err != nil {
+					msg.GetLogger().Println(getMessage("cannot load state entry:", err))
+					continue
 				}
-
-				hlog.Verbosef("instance %s skipped", v)
+				return &entry
 			}
 		}
-	}
+		return nil
 
-	return
+	case likely&likeFull != 0:
+		var likelyID hst.ID
+		if likelyID.UnmarshalText([]byte(name)) != nil {
+			return nil
+		}
+		msg.Verbose("argument looks like identifier")
+		for eh := range entries {
+			if eh.DecodeErr != nil {
+				msg.Verbose(getMessage("skipping instance:", eh.DecodeErr))
+				continue
+			}
+
+			if eh.ID == likelyID {
+				var entry hst.State
+				if _, err := eh.Load(&entry); err != nil {
+					msg.GetLogger().Println(getMessage("cannot load state entry:", err))
+					continue
+				}
+				return &entry
+			}
+		}
+		return nil
+
+	default:
+		panic("unreachable")
+	}
 }

cmd/hakurei/parse_test.go

@@ -0,0 +1,117 @@
+package main
+
+import (
+	"bytes"
+	"reflect"
+	"testing"
+	"time"
+
+	"hakurei.app/check"
+	"hakurei.app/hst"
+	"hakurei.app/internal/store"
+	"hakurei.app/message"
+)
+
+func TestShortIdentifier(t *testing.T) {
+	t.Parallel()
+	id := hst.ID{
+		0x01, 0x23, 0x45, 0x67, 0x89, 0xab, 0xcd, 0xef,
+		0xfe, 0xdc, 0xba, 0x98, 0x76, 0x54, 0x32, 0x10,
+	}
+
+	const want = "fedcba98"
+	if got := shortIdentifier(&id); got != want {
+		t.Errorf("shortIdentifier: %q, want %q", got, want)
+	}
+}
+
+func TestTryIdentifier(t *testing.T) {
+	t.Parallel()
+
+	msg := message.New(nil)
+	id := hst.ID{
+		0x01, 0x23, 0x45, 0x67, 0x89, 0xab, 0xcd, 0xef,
+		0xfe, 0xdc, 0xba, 0x98, 0x76, 0x54, 0x32, 0x10,
+	}
+	withBase := func(extra ...hst.State) []hst.State {
+		return append([]hst.State{
+			{ID: (hst.ID)(bytes.Repeat([]byte{0xaa}, len(hst.ID{}))), PID: 0xbeef, ShimPID: 0xcafe, Config: hst.Template(), Time: time.Unix(0, 0xdeadbeef0)},
+			{ID: (hst.ID)(bytes.Repeat([]byte{0xab}, len(hst.ID{}))), PID: 0x1beef, ShimPID: 0x1cafe, Config: hst.Template(), Time: time.Unix(0, 0xdeadbeef1)},
+			{ID: (hst.ID)(bytes.Repeat([]byte{0xf0}, len(hst.ID{}))), PID: 0x2beef, ShimPID: 0x2cafe, Config: hst.Template(), Time: time.Unix(0, 0xdeadbeef2)},
+
+			{ID: (hst.ID)(bytes.Repeat([]byte{0xfe}, len(hst.ID{}))), PID: 0xbed, ShimPID: 0xfff, Config: func() *hst.Config {
+				template := hst.Template()
+				template.Identity = hst.IdentityEnd
+				return template
+			}(), Time: time.Unix(0, 0xcafebabe0)},
+			{ID: (hst.ID)(bytes.Repeat([]byte{0xfc}, len(hst.ID{}))), PID: 0x1bed, ShimPID: 0x1fff, Config: func() *hst.Config {
+				template := hst.Template()
+				template.Identity = 0xfc
+				return template
+			}(), Time: time.Unix(0, 0xcafebabe1)},
+			{ID: (hst.ID)(bytes.Repeat([]byte{0xce}, len(hst.ID{}))), PID: 0x2bed, ShimPID: 0x2fff, Config: func() *hst.Config {
+				template := hst.Template()
+				template.Identity = 0xce
+				return template
+			}(), Time: time.Unix(0, 0xcafebabe2)},
+		}, extra...)
+	}
+	sampleEntry := hst.State{
+		ID:      id,
+		PID:     0xcafe,
+		ShimPID: 0xdead,
+		Config:  hst.Template(),
+	}
+
+	testCases := []struct {
+		name string
+		s    string
+		data []hst.State
+		want *hst.State
+	}{
+		{"likely entries fault", "ffffffff", nil, nil},
+
+		{"likely short too short", "ff", nil, nil},
+		{"likely short too long", "fffffffffffffffff", nil, nil},
+		{"likely short invalid lower", "fffffff\x00", nil, nil},
+		{"likely short invalid higher", "0000000\xff", nil, nil},
+		{"short no match", "fedcba98", withBase(), nil},
+		{"short match", "fedcba98", withBase(sampleEntry), &sampleEntry},
+		{"short match single", "fedcba98", []hst.State{sampleEntry}, &sampleEntry},
+		{"short match longer", "fedcba98765", withBase(sampleEntry), &sampleEntry},
+
+		{"likely long invalid", "0123456789abcdeffedcba987654321\x00", nil, nil},
+		{"long no match", "0123456789abcdeffedcba9876543210", withBase(), nil},
+		{"long match", "0123456789abcdeffedcba9876543210", withBase(sampleEntry), &sampleEntry},
+		{"long match single", "0123456789abcdeffedcba9876543210", []hst.State{sampleEntry}, &sampleEntry},
+	}
+	for _, tc := range testCases {
+		base := check.MustAbs(t.TempDir()).Append("store")
+		s := store.New(base)
+		for i := range tc.data {
+			if h, err := s.Handle(tc.data[i].Identity); err != nil {
+				t.Fatalf("Handle: error = %v", err)
+			} else {
+				var unlock func()
+				if unlock, err = h.Lock(); err != nil {
+					t.Fatalf("Lock: error = %v", err)
+				}
+				_, err = h.Save(&tc.data[i])
+				unlock()
+				if err != nil {
+					t.Fatalf("Save: error = %v", err)
+				}
+			}
+		}
+
+		// store must not be written to beyond this point
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			got := tryIdentifier(msg, tc.s, store.New(base))
+			if !reflect.DeepEqual(got, tc.want) {
+				t.Errorf("tryIdentifier: %#v, want %#v", got, tc.want)
+			}
+		})
+	}
+}

cmd/hakurei/print.go

@@ -1,11 +1,10 @@
 package main
 
 import (
-	"encoding/json"
+	"bytes"
 	"fmt"
 	"io"
 	"log"
-	"os"
 	"slices"
 	"strconv"
 	"strings"
@@ -13,46 +12,43 @@ import (
 	"time"
 
 	"hakurei.app/hst"
-	"hakurei.app/internal/app/state"
-	"hakurei.app/internal/hlog"
-	"hakurei.app/system/dbus"
+	"hakurei.app/internal/outcome"
+	"hakurei.app/internal/store"
+	"hakurei.app/message"
 )
 
+// printShowSystem populates and writes a representation of [hst.Info] to output.
 func printShowSystem(output io.Writer, short, flagJSON bool) {
 	t := newPrinter(output)
 	defer t.MustFlush()
-
-	info := &hst.Info{Paths: std.Paths()}
-
-	// get hid by querying uid of identity 0
-	if uid, err := std.Uid(0); err != nil {
-		hlog.PrintBaseError(err, "cannot obtain uid from setuid wrapper:")
-		os.Exit(1)
-	} else {
-		info.User = (uid / 10000) - 100
-	}
+	hi := outcome.Info()
 
 	if flagJSON {
-		printJSON(output, short, info)
+		encodeJSON(log.Fatal, output, short, hi)
 		return
 	}
 
-	t.Printf("User:\t%d\n", info.User)
-	t.Printf("TempDir:\t%s\n", info.TempDir)
-	t.Printf("SharePath:\t%s\n", info.SharePath)
-	t.Printf("RuntimePath:\t%s\n", info.RuntimePath)
-	t.Printf("RunDirPath:\t%s\n", info.RunDirPath)
+	t.Printf("Version:\t%s (libwayland %s)\n", hi.Version, hi.WaylandVersion)
+	t.Printf("User:\t%d\n", hi.User)
+	t.Printf("TempDir:\t%s\n", hi.TempDir)
+	t.Printf("SharePath:\t%s\n", hi.SharePath)
+	t.Printf("RuntimePath:\t%s\n", hi.RuntimePath)
+	t.Printf("RunDirPath:\t%s\n", hi.RunDirPath)
 }
 
+// printShowInstance writes a representation of [hst.State] or [hst.Config] to output.
 func printShowInstance(
 	output io.Writer, now time.Time,
-	instance *state.State, config *hst.Config,
-	short, flagJSON bool) {
+	instance *hst.State, config *hst.Config,
+	short, flagJSON bool,
+) (valid bool) {
+	valid = true
+
 	if flagJSON {
 		if instance != nil {
-			printJSON(output, short, instance)
+			encodeJSON(log.Fatal, output, short, instance)
 		} else {
-			printJSON(output, short, config)
+			encodeJSON(log.Fatal, output, short, config)
 		}
 		return
 	}
@@ -60,13 +56,21 @@ func printShowInstance(
 	t := newPrinter(output)
 	defer t.MustFlush()
 
-	if config.Container == nil {
-		mustPrint(output, "Warning: this configuration uses permissive defaults!\n\n")
+	if err := config.Validate(hst.VAllowInsecure); err != nil {
+		valid = false
+		if m, ok := message.GetMessage(err); ok {
+			mustPrint(output, "Error: "+m+"!\n\n")
+		}
+	}
+
+	if config == nil {
+		// nothing to print
+		return
 	}
 
 	if instance != nil {
 		t.Printf("State\n")
-		t.Printf(" Instance:\t%s (%d)\n", instance.ID.String(), instance.PID)
+		t.Printf(" Instance:\t%s (%d -> %d)\n", instance.ID.String(), instance.PID, instance.ShimPID)
 		t.Printf(" Uptime:\t%s\n", now.Sub(instance.Time).Round(time.Second).String())
 		t.Printf("\n")
 	}
@@ -81,39 +85,33 @@ func printShowInstance(
 	if len(config.Groups) > 0 {
 		t.Printf(" Groups:\t%s\n", strings.Join(config.Groups, ", "))
 	}
-	if config.Home != nil {
-		t.Printf(" Home:\t%s\n", config.Home)
-	}
 	if config.Container != nil {
-		params := config.Container
-		if params.Hostname != "" {
-			t.Printf(" Hostname:\t%s\n", params.Hostname)
-		}
-		flags := make([]string, 0, 7)
-		writeFlag := func(name string, value bool) {
-			if value {
-				flags = append(flags, name)
+		flags := config.Container.Flags.String()
+
+		// this is included in the upper hst.Config struct but is relevant here
+		const flagDirectWayland = "directwl"
+		if config.DirectWayland {
+			// hardcoded value when every flag is unset
+			if flags == "none" {
+				flags = flagDirectWayland
+			} else {
+				flags += ", " + flagDirectWayland
 			}
 		}
-		writeFlag("userns", params.Userns)
-		writeFlag("devel", params.Devel)
-		writeFlag("net", params.HostNet)
-		writeFlag("abstract", params.HostAbstract)
-		writeFlag("device", params.Device)
-		writeFlag("tty", params.Tty)
-		writeFlag("mapuid", params.MapRealUID)
-		writeFlag("directwl", config.DirectWayland)
-		if len(flags) == 0 {
-			flags = append(flags, "none")
-		}
-		t.Printf(" Flags:\t%s\n", strings.Join(flags, " "))
+		t.Printf(" Flags:\t%s\n", flags)
 
-		if config.Path != nil {
-			t.Printf(" Path:\t%s\n", config.Path)
+		if config.Container.Home != nil {
+			t.Printf(" Home:\t%s\n", config.Container.Home)
+		}
+		if config.Container.Hostname != "" {
+			t.Printf(" Hostname:\t%s\n", config.Container.Hostname)
+		}
+		if config.Container.Path != nil {
+			t.Printf(" Path:\t%s\n", config.Container.Path)
+		}
+		if len(config.Container.Args) > 0 {
+			t.Printf(" Arguments:\t%s\n", strings.Join(config.Container.Args, " "))
 		}
-	}
-	if len(config.Args) > 0 {
-		t.Printf(" Arguments:\t%s\n", strings.Join(config.Args, " "))
 	}
 	t.Printf("\n")
 
@@ -122,6 +120,7 @@ func printShowInstance(
 			t.Printf("Filesystem\n")
 			for _, f := range config.Container.Filesystem {
 				if !f.Valid() {
+					valid = false
 					t.Println(" <invalid>")
 					continue
 				}
@@ -131,17 +130,14 @@ func printShowInstance(
 		}
 		if len(config.ExtraPerms) > 0 {
 			t.Printf("Extra ACL\n")
-			for _, p := range config.ExtraPerms {
-				if p == nil {
-					continue
-				}
-				t.Printf(" %s\n", p.String())
+			for i := range config.ExtraPerms {
+				t.Printf(" %s\n", config.ExtraPerms[i].String())
 			}
 			t.Printf("\n")
 		}
 	}
 
-	printDBus := func(c *dbus.Config) {
+	printDBus := func(c *hst.BusConfig) {
 		t.Printf(" Filter:\t%v\n", c.Filter)
 		if len(c.See) > 0 {
 			t.Printf(" See:\t%q\n", c.See)
@@ -169,59 +165,57 @@ func printShowInstance(
 		printDBus(config.SystemBus)
 		t.Printf("\n")
 	}
+
+	return
 }
 
-func printPs(output io.Writer, now time.Time, s state.Store, short, flagJSON bool) {
-	var entries state.Entries
-	if e, err := state.Join(s); err != nil {
-		log.Fatalf("cannot join store: %v", err)
-	} else {
-		entries = e
-	}
-	if err := s.Close(); err != nil {
-		log.Printf("cannot close store: %v", err)
+// printPs writes a representation of active instances to output.
+func printPs(msg message.Msg, output io.Writer, now time.Time, s *store.Store, short, flagJSON bool) {
+	f := func(a func(eh *store.EntryHandle)) {
+		entries, copyError := s.All()
+		for eh := range entries {
+			a(eh)
+		}
+		if err := copyError(); err != nil {
+			msg.GetLogger().Println(getMessage("cannot iterate over store:", err))
+		}
 	}
 
-	if !short && flagJSON {
-		es := make(map[string]*state.State, len(entries))
-		for id, instance := range entries {
-			es[id.String()] = instance
+	if short { // short output requires identifier only
+		var identifiers []*hst.ID
+		f(func(eh *store.EntryHandle) {
+			if _, err := eh.Load(nil); err != nil { // passes through decode error
+				msg.GetLogger().Println(getMessage("cannot validate state entry header:", err))
+				return
+			}
+			identifiers = append(identifiers, &eh.ID)
+		})
+		slices.SortFunc(identifiers, func(a, b *hst.ID) int { return bytes.Compare(a[:], b[:]) })
+
+		if flagJSON {
+			encodeJSON(log.Fatal, output, short, identifiers)
+		} else {
+			for _, id := range identifiers {
+				mustPrintln(output, shortIdentifier(id))
+			}
 		}
-		printJSON(output, short, es)
 		return
 	}
 
-	// sort state entries by id string to ensure consistency between runs
-	exp := make([]*expandedStateEntry, 0, len(entries))
-	for id, instance := range entries {
-		// gracefully skip nil states
-		if instance == nil {
-			log.Printf("got invalid state entry %s", id.String())
-			continue
+	// long output requires full instance state
+	var instances []*hst.State
+	f(func(eh *store.EntryHandle) {
+		var state hst.State
+		if _, err := eh.Load(&state); err != nil { // passes through decode error
+			msg.GetLogger().Println(getMessage("cannot load state entry:", err))
+			return
 		}
+		instances = append(instances, &state)
+	})
+	slices.SortFunc(instances, func(a, b *hst.State) int { return bytes.Compare(a.ID[:], b.ID[:]) })
 
-		// gracefully skip inconsistent states
-		if id != instance.ID {
-			log.Printf("possible store corruption: entry %s has id %s",
-				id.String(), instance.ID.String())
-			continue
-		}
-		exp = append(exp, &expandedStateEntry{s: id.String(), State: instance})
-	}
-	slices.SortFunc(exp, func(a, b *expandedStateEntry) int { return a.Time.Compare(b.Time) })
-
-	if short {
-		if flagJSON {
-			v := make([]string, len(exp))
-			for i, e := range exp {
-				v[i] = e.s
-			}
-			printJSON(output, short, v)
-		} else {
-			for _, e := range exp {
-				mustPrintln(output, e.s[:8])
-			}
-		}
+	if flagJSON {
+		encodeJSON(log.Fatal, output, short, instances)
 		return
 	}
 
@@ -229,61 +223,48 @@ func printPs(output io.Writer, now time.Time, s state.Store, short, flagJSON boo
 	defer t.MustFlush()
 
 	t.Println("\tInstance\tPID\tApplication\tUptime")
-	for _, e := range exp {
-		if len(e.s) != 1<<5 {
-			// unreachable
-			log.Printf("possible store corruption: invalid instance string %s", e.s)
-			continue
-		}
-
+	for _, instance := range instances {
 		as := "(No configuration information)"
-		if e.Config != nil {
-			as = strconv.Itoa(e.Config.Identity)
-			id := e.Config.ID
+		if instance.Config != nil {
+			as = strconv.Itoa(instance.Config.Identity)
+			id := instance.Config.ID
 			if id == "" {
-				id = "app.hakurei." + e.s[:8]
+				id = "app.hakurei." + shortIdentifier(&instance.ID)
 			}
 			as += " (" + id + ")"
 		}
 		t.Printf("\t%s\t%d\t%s\t%s\n",
-			e.s[:8], e.PID, as, now.Sub(e.Time).Round(time.Second).String())
-	}
-}
-
-type expandedStateEntry struct {
-	s string
-	*state.State
-}
-
-func printJSON(output io.Writer, short bool, v any) {
-	encoder := json.NewEncoder(output)
-	if !short {
-		encoder.SetIndent("", "  ")
-	}
-	if err := encoder.Encode(v); err != nil {
-		log.Fatalf("cannot serialise: %v", err)
+			shortIdentifier(&instance.ID), instance.PID, as, now.Sub(instance.Time).Round(time.Second).String())
 	}
 }
 
+// newPrinter returns a configured, wrapped [tabwriter.Writer].
 func newPrinter(output io.Writer) *tp { return &tp{tabwriter.NewWriter(output, 0, 1, 4, ' ', 0)} }
 
+// tp wraps [tabwriter.Writer] to provide additional formatting methods.
 type tp struct{ *tabwriter.Writer }
 
+// Printf calls [fmt.Fprintf] on the underlying [tabwriter.Writer].
 func (p *tp) Printf(format string, a ...any) {
 	if _, err := fmt.Fprintf(p, format, a...); err != nil {
 		log.Fatalf("cannot write to tabwriter: %v", err)
 	}
 }
+
+// Println calls [fmt.Fprintln] on the underlying [tabwriter.Writer].
 func (p *tp) Println(a ...any) {
 	if _, err := fmt.Fprintln(p, a...); err != nil {
 		log.Fatalf("cannot write to tabwriter: %v", err)
 	}
 }
+
+// MustFlush calls the Flush method of [tabwriter.Writer] and calls [log.Fatalf] on a non-nil error.
 func (p *tp) MustFlush() {
 	if err := p.Writer.Flush(); err != nil {
 		log.Fatalf("cannot flush tabwriter: %v", err)
 	}
 }
+
 func mustPrint(output io.Writer, a ...any) {
 	if _, err := fmt.Fprint(output, a...); err != nil {
 		log.Fatalf("cannot print: %v", err)
@@ -294,3 +275,11 @@ func mustPrintln(output io.Writer, a ...any) {
 		log.Fatalf("cannot print: %v", err)
 	}
 }
+
+// getMessage returns a [message.Error] message if available, or err prefixed with fallback otherwise.
+func getMessage(fallback string, err error) string {
+	if m, ok := message.GetMessage(err); ok {
+		return m
+	}
+	return fmt.Sprintln(fallback, err)
+}

cmd/hpkg/README

@@ -1,7 +0,0 @@
-This program is a proof of concept and is now deprecated. It is only kept
-around for API demonstration purposes and to make the most out of the test
-suite.
-
-This program is replaced by planterette, which can be found at
-https://git.gensokyo.uk/security/planterette. Development effort should be
-focused there instead.

cmd/hpkg/app.go

@@ -1,161 +0,0 @@
-package main
-
-import (
-	"encoding/json"
-	"log"
-	"os"
-
-	"hakurei.app/container"
-	"hakurei.app/container/seccomp"
-	"hakurei.app/hst"
-	"hakurei.app/system/dbus"
-)
-
-type appInfo struct {
-	Name    string `json:"name"`
-	Version string `json:"version"`
-
-	// passed through to [hst.Config]
-	ID string `json:"id"`
-	// passed through to [hst.Config]
-	Identity int `json:"identity"`
-	// passed through to [hst.Config]
-	Groups []string `json:"groups,omitempty"`
-	// passed through to [hst.Config]
-	Devel bool `json:"devel,omitempty"`
-	// passed through to [hst.Config]
-	Userns bool `json:"userns,omitempty"`
-	// passed through to [hst.Config]
-	HostNet bool `json:"net,omitempty"`
-	// passed through to [hst.Config]
-	HostAbstract bool `json:"abstract,omitempty"`
-	// passed through to [hst.Config]
-	Device bool `json:"dev,omitempty"`
-	// passed through to [hst.Config]
-	Tty bool `json:"tty,omitempty"`
-	// passed through to [hst.Config]
-	MapRealUID bool `json:"map_real_uid,omitempty"`
-	// passed through to [hst.Config]
-	DirectWayland bool `json:"direct_wayland,omitempty"`
-	// passed through to [hst.Config]
-	SystemBus *dbus.Config `json:"system_bus,omitempty"`
-	// passed through to [hst.Config]
-	SessionBus *dbus.Config `json:"session_bus,omitempty"`
-	// passed through to [hst.Config]
-	Enablements *hst.Enablements `json:"enablements,omitempty"`
-
-	// passed through to [hst.Config]
-	Multiarch bool `json:"multiarch,omitempty"`
-	// passed through to [hst.Config]
-	Bluetooth bool `json:"bluetooth,omitempty"`
-
-	// allow gpu access within sandbox
-	GPU bool `json:"gpu"`
-	// store path to nixGL mesa wrappers
-	Mesa string `json:"mesa,omitempty"`
-	// store path to nixGL source
-	NixGL string `json:"nix_gl,omitempty"`
-	// store path to activate-and-exec script
-	Launcher *container.Absolute `json:"launcher"`
-	// store path to /run/current-system
-	CurrentSystem *container.Absolute `json:"current_system"`
-	// store path to home-manager activation package
-	ActivationPackage string `json:"activation_package"`
-}
-
-func (app *appInfo) toHst(pathSet *appPathSet, pathname *container.Absolute, argv []string, flagDropShell bool) *hst.Config {
-	config := &hst.Config{
-		ID: app.ID,
-
-		Path: pathname,
-		Args: argv,
-
-		Enablements: app.Enablements,
-
-		SystemBus:     app.SystemBus,
-		SessionBus:    app.SessionBus,
-		DirectWayland: app.DirectWayland,
-
-		Username: "hakurei",
-		Shell:    pathShell,
-		Home:     pathDataData.Append(app.ID),
-
-		Identity: app.Identity,
-		Groups:   app.Groups,
-
-		Container: &hst.ContainerConfig{
-			Hostname:     formatHostname(app.Name),
-			Devel:        app.Devel,
-			Userns:       app.Userns,
-			HostNet:      app.HostNet,
-			HostAbstract: app.HostAbstract,
-			Device:       app.Device,
-			Tty:          app.Tty || flagDropShell,
-			MapRealUID:   app.MapRealUID,
-			Filesystem: []hst.FilesystemConfigJSON{
-				{FilesystemConfig: &hst.FSBind{Target: container.AbsFHSEtc, Source: pathSet.cacheDir.Append("etc"), Special: true}},
-				{FilesystemConfig: &hst.FSBind{Source: pathSet.nixPath.Append("store"), Target: pathNixStore}},
-				{FilesystemConfig: &hst.FSLink{Target: pathCurrentSystem, Linkname: app.CurrentSystem.String()}},
-				{FilesystemConfig: &hst.FSLink{Target: pathBin, Linkname: pathSwBin.String()}},
-				{FilesystemConfig: &hst.FSLink{Target: container.AbsFHSUsrBin, Linkname: pathSwBin.String()}},
-				{FilesystemConfig: &hst.FSBind{Source: pathSet.metaPath, Target: hst.AbsTmp.Append("app")}},
-				{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSEtc.Append("resolv.conf"), Optional: true}},
-				{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSSys.Append("block"), Optional: true}},
-				{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSSys.Append("bus"), Optional: true}},
-				{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSSys.Append("class"), Optional: true}},
-				{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSSys.Append("dev"), Optional: true}},
-				{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSSys.Append("devices"), Optional: true}},
-				{FilesystemConfig: &hst.FSBind{Target: pathDataData.Append(app.ID), Source: pathSet.homeDir, Write: true, Ensure: true}},
-			},
-		},
-		ExtraPerms: []*hst.ExtraPermConfig{
-			{Path: dataHome, Execute: true},
-			{Ensure: true, Path: pathSet.baseDir, Read: true, Write: true, Execute: true},
-		},
-	}
-	if app.Multiarch {
-		config.Container.SeccompFlags |= seccomp.AllowMultiarch
-	}
-	if app.Bluetooth {
-		config.Container.SeccompFlags |= seccomp.AllowBluetooth
-	}
-	return config
-}
-
-func loadAppInfo(name string, beforeFail func()) *appInfo {
-	bundle := new(appInfo)
-	if f, err := os.Open(name); err != nil {
-		beforeFail()
-		log.Fatalf("cannot open bundle: %v", err)
-	} else if err = json.NewDecoder(f).Decode(&bundle); err != nil {
-		beforeFail()
-		log.Fatalf("cannot parse bundle metadata: %v", err)
-	} else if err = f.Close(); err != nil {
-		log.Printf("cannot close bundle metadata: %v", err)
-		// not fatal
-	}
-
-	if bundle.ID == "" {
-		beforeFail()
-		log.Fatal("application identifier must not be empty")
-	}
-	if bundle.Launcher == nil {
-		beforeFail()
-		log.Fatal("launcher must not be empty")
-	}
-	if bundle.CurrentSystem == nil {
-		beforeFail()
-		log.Fatal("current-system must not be empty")
-	}
-
-	return bundle
-}
-
-func formatHostname(name string) string {
-	if h, err := os.Hostname(); err != nil {
-		log.Printf("cannot get hostname: %v", err)
-		return "hakurei-" + name
-	} else {
-		return h + "-" + name
-	}
-}

cmd/hpkg/build.nix

@@ -1,256 +0,0 @@
-{
-  nixpkgsFor,
-  system,
-  nixpkgs,
-  home-manager,
-}:
-
-{
-  lib,
-  stdenv,
-  closureInfo,
-  writeScript,
-  runtimeShell,
-  writeText,
-  symlinkJoin,
-  vmTools,
-  runCommand,
-  fetchFromGitHub,
-
-  zstd,
-  nix,
-  sqlite,
-
-  name ? throw "name is required",
-  version ? throw "version is required",
-  pname ? "${name}-${version}",
-  modules ? [ ],
-  nixosModules ? [ ],
-  script ? ''
-    exec "$SHELL" "$@"
-  '',
-
-  id ? name,
-  identity ? throw "identity is required",
-  groups ? [ ],
-  userns ? false,
-  net ? true,
-  dev ? false,
-  no_new_session ? false,
-  map_real_uid ? false,
-  direct_wayland ? false,
-  system_bus ? null,
-  session_bus ? null,
-
-  allow_wayland ? true,
-  allow_x11 ? false,
-  allow_dbus ? true,
-  allow_pulse ? true,
-  gpu ? allow_wayland || allow_x11,
-}:
-
-let
-  inherit (lib) optionals;
-
-  homeManagerConfiguration = home-manager.lib.homeManagerConfiguration {
-    pkgs = nixpkgsFor.${system};
-    modules = modules ++ [
-      {
-        home = {
-          username = "hakurei";
-          homeDirectory = "/data/data/${id}";
-          stateVersion = "22.11";
-        };
-      }
-    ];
-  };
-
-  launcher = writeScript "hakurei-${pname}" ''
-    #!${runtimeShell} -el
-    ${script}
-  '';
-
-  extraNixOSConfig =
-    { pkgs, ... }:
-    {
-      environment = {
-        etc.nixpkgs.source = nixpkgs.outPath;
-        systemPackages = [ pkgs.nix ];
-      };
-
-      imports = nixosModules;
-    };
-  nixos = nixpkgs.lib.nixosSystem {
-    inherit system;
-    modules = [
-      extraNixOSConfig
-      { nix.settings.experimental-features = [ "flakes" ]; }
-      { nix.settings.experimental-features = [ "nix-command" ]; }
-      { boot.isContainer = true; }
-      { system.stateVersion = "22.11"; }
-    ];
-  };
-
-  etc = vmTools.runInLinuxVM (
-    runCommand "etc" { } ''
-      mkdir -p /etc
-      ${nixos.config.system.build.etcActivationCommands}
-
-      # remove unused files
-      rm -rf /etc/sudoers
-
-      mkdir -p $out
-      tar -C /etc -cf "$out/etc.tar" .
-    ''
-  );
-
-  extendSessionDefault = id: ext: {
-    filter = true;
-
-    talk = [ "org.freedesktop.Notifications" ] ++ ext.talk;
-    own =
-      (optionals (id != null) [
-        "${id}.*"
-        "org.mpris.MediaPlayer2.${id}.*"
-      ])
-      ++ ext.own;
-
-    inherit (ext) call broadcast;
-  };
-
-  nixGL = fetchFromGitHub {
-    owner = "nix-community";
-    repo = "nixGL";
-    rev = "310f8e49a149e4c9ea52f1adf70cdc768ec53f8a";
-    hash = "sha256-lnzZQYG0+EXl/6NkGpyIz+FEOc/DSEG57AP1VsdeNrM=";
-  };
-
-  mesaWrappers =
-    let
-      isIntelX86Platform = system == "x86_64-linux";
-      nixGLPackages = import (nixGL + "/default.nix") {
-        pkgs = nixpkgs.legacyPackages.${system};
-        enable32bits = isIntelX86Platform;
-        enableIntelX86Extensions = isIntelX86Platform;
-      };
-    in
-    symlinkJoin {
-      name = "nixGL-mesa";
-      paths = with nixGLPackages; [
-        nixGLIntel
-        nixVulkanIntel
-      ];
-    };
-
-  info = builtins.toJSON {
-    inherit
-      name
-      version
-      id
-      identity
-      launcher
-      groups
-      userns
-      net
-      dev
-      no_new_session
-      map_real_uid
-      direct_wayland
-      system_bus
-      gpu
-      ;
-
-    session_bus =
-      if session_bus != null then
-        (session_bus (extendSessionDefault id))
-      else
-        (extendSessionDefault id {
-          talk = [ ];
-          own = [ ];
-          call = { };
-          broadcast = { };
-        });
-
-    enablements = {
-      wayland = allow_wayland;
-      x11 = allow_x11;
-      dbus = allow_dbus;
-      pulse = allow_pulse;
-    };
-
-    mesa = if gpu then mesaWrappers else null;
-    nix_gl = if gpu then nixGL else null;
-    current_system = nixos.config.system.build.toplevel;
-    activation_package = homeManagerConfiguration.activationPackage;
-  };
-in
-
-stdenv.mkDerivation {
-  name = "${pname}.pkg";
-  inherit version;
-  __structuredAttrs = true;
-
-  nativeBuildInputs = [
-    zstd
-    nix
-    sqlite
-  ];
-
-  buildCommand = ''
-    NIX_ROOT="$(mktemp -d)"
-    export USER="nobody"
-
-    # create bootstrap store
-    bootstrapClosureInfo="${
-      closureInfo {
-        rootPaths = [
-          nix
-          nixos.config.system.build.toplevel
-        ];
-      }
-    }"
-    echo "copying bootstrap store paths..."
-    mkdir -p "$NIX_ROOT/nix/store"
-    xargs -n 1 -a "$bootstrapClosureInfo/store-paths" cp -at "$NIX_ROOT/nix/store/"
-    NIX_REMOTE="local?root=$NIX_ROOT" nix-store --load-db < "$bootstrapClosureInfo/registration"
-    NIX_REMOTE="local?root=$NIX_ROOT" nix-store --optimise
-    sqlite3 "$NIX_ROOT/nix/var/nix/db/db.sqlite" "UPDATE ValidPaths SET registrationTime = ''${SOURCE_DATE_EPOCH}"
-    chmod -R +r "$NIX_ROOT/nix/var"
-
-    # create binary cache
-    closureInfo="${
-      closureInfo {
-        rootPaths = [
-          homeManagerConfiguration.activationPackage
-          launcher
-        ]
-        ++ optionals gpu [
-          mesaWrappers
-          nixGL
-        ];
-      }
-    }"
-    echo "copying application paths..."
-    TMP_STORE="$(mktemp -d)"
-    mkdir -p "$TMP_STORE/nix/store"
-    xargs -n 1 -a "$closureInfo/store-paths" cp -at "$TMP_STORE/nix/store/"
-    NIX_REMOTE="local?root=$TMP_STORE" nix-store --load-db < "$closureInfo/registration"
-    sqlite3 "$TMP_STORE/nix/var/nix/db/db.sqlite" "UPDATE ValidPaths SET registrationTime = ''${SOURCE_DATE_EPOCH}"
-    NIX_REMOTE="local?root=$TMP_STORE" nix --offline --extra-experimental-features nix-command \
-        --verbose --log-format raw-with-logs \
-        copy --all --no-check-sigs --to \
-        "file://$NIX_ROOT/res?compression=zstd&compression-level=19&parallel-compression=true"
-
-    # package /etc
-    mkdir -p "$NIX_ROOT/etc"
-    tar -C "$NIX_ROOT/etc" -xf "${etc}/etc.tar"
-
-    # write metadata
-    cp "${writeText "bundle.json" info}" "$NIX_ROOT/bundle.json"
-
-    # create an intermediate file to improve zstd performance
-    INTER="$(mktemp)"
-    tar -C "$NIX_ROOT" -cf "$INTER" .
-    zstd -T0 -19 -fo "$out" "$INTER"
-  '';
-}

cmd/hpkg/main.go

@@ -1,334 +0,0 @@
-package main
-
-import (
-	"context"
-	"encoding/json"
-	"errors"
-	"log"
-	"os"
-	"os/signal"
-	"path"
-	"syscall"
-
-	"hakurei.app/command"
-	"hakurei.app/container"
-	"hakurei.app/hst"
-	"hakurei.app/internal"
-	"hakurei.app/internal/hlog"
-)
-
-var (
-	errSuccess = errors.New("success")
-)
-
-func init() {
-	hlog.Prepare("hpkg")
-	if err := os.Setenv("SHELL", pathShell.String()); err != nil {
-		log.Fatalf("cannot set $SHELL: %v", err)
-	}
-}
-
-func main() {
-	if os.Geteuid() == 0 {
-		log.Fatal("this program must not run as root")
-	}
-
-	ctx, stop := signal.NotifyContext(context.Background(),
-		syscall.SIGINT, syscall.SIGTERM)
-	defer stop() // unreachable
-
-	var (
-		flagVerbose   bool
-		flagDropShell bool
-	)
-	c := command.New(os.Stderr, log.Printf, "hpkg", func([]string) error { internal.InstallOutput(flagVerbose); return nil }).
-		Flag(&flagVerbose, "v", command.BoolFlag(false), "Print debug messages to the console").
-		Flag(&flagDropShell, "s", command.BoolFlag(false), "Drop to a shell in place of next hakurei action")
-
-	{
-		var (
-			flagDropShellActivate bool
-		)
-		c.NewCommand("install", "Install an application from its package", func(args []string) error {
-			if len(args) != 1 {
-				log.Println("invalid argument")
-				return syscall.EINVAL
-			}
-			pkgPath := args[0]
-			if !path.IsAbs(pkgPath) {
-				if dir, err := os.Getwd(); err != nil {
-					log.Printf("cannot get current directory: %v", err)
-					return err
-				} else {
-					pkgPath = path.Join(dir, pkgPath)
-				}
-			}
-
-			/*
-				Look up paths to programs started by hpkg.
-				This is done here to ease error handling as cleanup is not yet required.
-			*/
-
-			var (
-				_     = lookPath("zstd")
-				tar   = lookPath("tar")
-				chmod = lookPath("chmod")
-				rm    = lookPath("rm")
-			)
-
-			/*
-				Extract package and set up for cleanup.
-			*/
-
-			var workDir *container.Absolute
-			if p, err := os.MkdirTemp("", "hpkg.*"); err != nil {
-				log.Printf("cannot create temporary directory: %v", err)
-				return err
-			} else if workDir, err = container.NewAbs(p); err != nil {
-				log.Printf("invalid temporary directory: %v", err)
-				return err
-			}
-			cleanup := func() {
-				// should be faster than a native implementation
-				mustRun(chmod, "-R", "+w", workDir.String())
-				mustRun(rm, "-rf", workDir.String())
-			}
-			beforeRunFail.Store(&cleanup)
-
-			mustRun(tar, "-C", workDir.String(), "-xf", pkgPath)
-
-			/*
-				Parse bundle and app metadata, do pre-install checks.
-			*/
-
-			bundle := loadAppInfo(path.Join(workDir.String(), "bundle.json"), cleanup)
-			pathSet := pathSetByApp(bundle.ID)
-
-			a := bundle
-			if s, err := os.Stat(pathSet.metaPath.String()); err != nil {
-				if !os.IsNotExist(err) {
-					cleanup()
-					log.Printf("cannot access %q: %v", pathSet.metaPath, err)
-					return err
-				}
-				// did not modify app, clean installation condition met later
-			} else if s.IsDir() {
-				cleanup()
-				log.Printf("metadata path %q is not a file", pathSet.metaPath)
-				return syscall.EBADMSG
-			} else {
-				a = loadAppInfo(pathSet.metaPath.String(), cleanup)
-				if a.ID != bundle.ID {
-					cleanup()
-					log.Printf("app %q claims to have identifier %q",
-						bundle.ID, a.ID)
-					return syscall.EBADE
-				}
-				// sec: should verify credentials
-			}
-
-			if a != bundle {
-				// do not try to re-install
-				if a.NixGL == bundle.NixGL &&
-					a.CurrentSystem == bundle.CurrentSystem &&
-					a.Launcher == bundle.Launcher &&
-					a.ActivationPackage == bundle.ActivationPackage {
-					cleanup()
-					log.Printf("package %q is identical to local application %q",
-						pkgPath, a.ID)
-					return errSuccess
-				}
-
-				// identity determines uid
-				if a.Identity != bundle.Identity {
-					cleanup()
-					log.Printf("package %q identity %d differs from installed %d",
-						pkgPath, bundle.Identity, a.Identity)
-					return syscall.EBADE
-				}
-
-				// sec: should compare version string
-				hlog.Verbosef("installing application %q version %q over local %q",
-					bundle.ID, bundle.Version, a.Version)
-			} else {
-				hlog.Verbosef("application %q clean installation", bundle.ID)
-				// sec: should install credentials
-			}
-
-			/*
-				Setup steps for files owned by the target user.
-			*/
-
-			withCacheDir(ctx, "install", []string{
-				// export inner bundle path in the environment
-				"export BUNDLE=" + hst.Tmp + "/bundle",
-				// replace inner /etc
-				"mkdir -p etc",
-				"chmod -R +w etc",
-				"rm -rf etc",
-				"cp -dRf $BUNDLE/etc etc",
-				// replace inner /nix
-				"mkdir -p nix",
-				"chmod -R +w nix",
-				"rm -rf nix",
-				"cp -dRf /nix nix",
-				// copy from binary cache
-				"nix copy --offline --no-check-sigs --all --from file://$BUNDLE/res --to $PWD",
-				// deduplicate nix store
-				"nix store --offline --store $PWD optimise",
-				// make cache directory world-readable for autoetc
-				"chmod 0755 .",
-			}, workDir, bundle, pathSet, flagDropShell, cleanup)
-
-			if bundle.GPU {
-				withCacheDir(ctx, "mesa-wrappers", []string{
-					// link nixGL mesa wrappers
-					"mkdir -p nix/.nixGL",
-					"ln -s " + bundle.Mesa + "/bin/nixGLIntel nix/.nixGL/nixGL",
-					"ln -s " + bundle.Mesa + "/bin/nixVulkanIntel nix/.nixGL/nixVulkan",
-				}, workDir, bundle, pathSet, false, cleanup)
-			}
-
-			/*
-				Activate home-manager generation.
-			*/
-
-			withNixDaemon(ctx, "activate", []string{
-				// clean up broken links
-				"mkdir -p .local/state/{nix,home-manager}",
-				"chmod -R +w .local/state/{nix,home-manager}",
-				"rm -rf .local/state/{nix,home-manager}",
-				// run activation script
-				bundle.ActivationPackage + "/activate",
-			}, false, func(config *hst.Config) *hst.Config { return config },
-				bundle, pathSet, flagDropShellActivate, cleanup)
-
-			/*
-				Installation complete. Write metadata to block re-installs or downgrades.
-			*/
-
-			// serialise metadata to ensure consistency
-			if f, err := os.OpenFile(pathSet.metaPath.String()+"~", os.O_CREATE|os.O_WRONLY|os.O_TRUNC, 0644); err != nil {
-				cleanup()
-				log.Printf("cannot create metadata file: %v", err)
-				return err
-			} else if err = json.NewEncoder(f).Encode(bundle); err != nil {
-				cleanup()
-				log.Printf("cannot write metadata: %v", err)
-				return err
-			} else if err = f.Close(); err != nil {
-				log.Printf("cannot close metadata file: %v", err)
-				// not fatal
-			}
-
-			if err := os.Rename(pathSet.metaPath.String()+"~", pathSet.metaPath.String()); err != nil {
-				cleanup()
-				log.Printf("cannot rename metadata file: %v", err)
-				return err
-			}
-
-			cleanup()
-			return errSuccess
-		}).
-			Flag(&flagDropShellActivate, "s", command.BoolFlag(false), "Drop to a shell on activation")
-	}
-
-	{
-		var (
-			flagDropShellNixGL bool
-			flagAutoDrivers    bool
-		)
-		c.NewCommand("start", "Start an application", func(args []string) error {
-			if len(args) < 1 {
-				log.Println("invalid argument")
-				return syscall.EINVAL
-			}
-
-			/*
-				Parse app metadata.
-			*/
-
-			id := args[0]
-			pathSet := pathSetByApp(id)
-			a := loadAppInfo(pathSet.metaPath.String(), func() {})
-			if a.ID != id {
-				log.Printf("app %q claims to have identifier %q", id, a.ID)
-				return syscall.EBADE
-			}
-
-			/*
-				Prepare nixGL.
-			*/
-
-			if a.GPU && flagAutoDrivers {
-				withNixDaemon(ctx, "nix-gl", []string{
-					"mkdir -p /nix/.nixGL/auto",
-					"rm -rf /nix/.nixGL/auto",
-					"export NIXPKGS_ALLOW_UNFREE=1",
-					"nix build --impure " +
-						"--out-link /nix/.nixGL/auto/opengl " +
-						"--override-input nixpkgs path:/etc/nixpkgs " +
-						"path:" + a.NixGL,
-					"nix build --impure " +
-						"--out-link /nix/.nixGL/auto/vulkan " +
-						"--override-input nixpkgs path:/etc/nixpkgs " +
-						"path:" + a.NixGL + "#nixVulkanNvidia",
-				}, true, func(config *hst.Config) *hst.Config {
-					config.Container.Filesystem = append(config.Container.Filesystem, []hst.FilesystemConfigJSON{
-						{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSEtc.Append("resolv.conf"), Optional: true}},
-						{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSSys.Append("block"), Optional: true}},
-						{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSSys.Append("bus"), Optional: true}},
-						{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSSys.Append("class"), Optional: true}},
-						{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSSys.Append("dev"), Optional: true}},
-						{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSSys.Append("devices"), Optional: true}},
-					}...)
-					appendGPUFilesystem(config)
-					return config
-				}, a, pathSet, flagDropShellNixGL, func() {})
-			}
-
-			/*
-				Create app configuration.
-			*/
-
-			pathname := a.Launcher
-			argv := make([]string, 1, len(args))
-			if flagDropShell {
-				pathname = pathShell
-				argv[0] = bash
-			} else {
-				argv[0] = a.Launcher.String()
-			}
-			argv = append(argv, args[1:]...)
-			config := a.toHst(pathSet, pathname, argv, flagDropShell)
-
-			/*
-				Expose GPU devices.
-			*/
-
-			if a.GPU {
-				config.Container.Filesystem = append(config.Container.Filesystem,
-					hst.FilesystemConfigJSON{FilesystemConfig: &hst.FSBind{Source: pathSet.nixPath.Append(".nixGL"), Target: hst.AbsTmp.Append("nixGL")}})
-				appendGPUFilesystem(config)
-			}
-
-			/*
-				Spawn app.
-			*/
-
-			mustRunApp(ctx, config, func() {})
-			return errSuccess
-		}).
-			Flag(&flagDropShellNixGL, "s", command.BoolFlag(false), "Drop to a shell on nixGL build").
-			Flag(&flagAutoDrivers, "auto-drivers", command.BoolFlag(false), "Attempt automatic opengl driver detection")
-	}
-
-	c.MustParse(os.Args[1:], func(err error) {
-		hlog.Verbosef("command returned %v", err)
-		if errors.Is(err, errSuccess) {
-			hlog.BeforeExit()
-			os.Exit(0)
-		}
-	})
-	log.Fatal("unreachable")
-}

cmd/hpkg/paths.go

@@ -1,116 +0,0 @@
-package main
-
-import (
-	"log"
-	"os"
-	"os/exec"
-	"strconv"
-	"sync/atomic"
-
-	"hakurei.app/container"
-	"hakurei.app/hst"
-	"hakurei.app/internal/hlog"
-)
-
-const bash = "bash"
-
-var (
-	dataHome *container.Absolute
-)
-
-func init() {
-	// dataHome
-	if a, err := container.NewAbs(os.Getenv("HAKUREI_DATA_HOME")); err == nil {
-		dataHome = a
-	} else {
-		dataHome = container.AbsFHSVarLib.Append("hakurei/" + strconv.Itoa(os.Getuid()))
-	}
-}
-
-var (
-	pathBin = container.AbsFHSRoot.Append("bin")
-
-	pathNix           = container.MustAbs("/nix/")
-	pathNixStore      = pathNix.Append("store/")
-	pathCurrentSystem = container.AbsFHSRun.Append("current-system")
-	pathSwBin         = pathCurrentSystem.Append("sw/bin/")
-	pathShell         = pathSwBin.Append(bash)
-
-	pathData     = container.MustAbs("/data")
-	pathDataData = pathData.Append("data")
-)
-
-func lookPath(file string) string {
-	if p, err := exec.LookPath(file); err != nil {
-		log.Fatalf("%s: command not found", file)
-		return ""
-	} else {
-		return p
-	}
-}
-
-var beforeRunFail = new(atomic.Pointer[func()])
-
-func mustRun(name string, arg ...string) {
-	hlog.Verbosef("spawning process: %q %q", name, arg)
-	cmd := exec.Command(name, arg...)
-	cmd.Stdin, cmd.Stdout, cmd.Stderr = os.Stdin, os.Stdout, os.Stderr
-	if err := cmd.Run(); err != nil {
-		if f := beforeRunFail.Swap(nil); f != nil {
-			(*f)()
-		}
-		log.Fatalf("%s: %v", name, err)
-	}
-}
-
-type appPathSet struct {
-	// ${dataHome}/${id}
-	baseDir *container.Absolute
-	// ${baseDir}/app
-	metaPath *container.Absolute
-	// ${baseDir}/files
-	homeDir *container.Absolute
-	// ${baseDir}/cache
-	cacheDir *container.Absolute
-	// ${baseDir}/cache/nix
-	nixPath *container.Absolute
-}
-
-func pathSetByApp(id string) *appPathSet {
-	pathSet := new(appPathSet)
-	pathSet.baseDir = dataHome.Append(id)
-	pathSet.metaPath = pathSet.baseDir.Append("app")
-	pathSet.homeDir = pathSet.baseDir.Append("files")
-	pathSet.cacheDir = pathSet.baseDir.Append("cache")
-	pathSet.nixPath = pathSet.cacheDir.Append("nix")
-	return pathSet
-}
-
-func appendGPUFilesystem(config *hst.Config) {
-	config.Container.Filesystem = append(config.Container.Filesystem, []hst.FilesystemConfigJSON{
-		// flatpak commit 763a686d874dd668f0236f911de00b80766ffe79
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("dri"), Device: true, Optional: true}},
-		// mali
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("mali"), Device: true, Optional: true}},
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("mali0"), Device: true, Optional: true}},
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("umplock"), Device: true, Optional: true}},
-		// nvidia
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidiactl"), Device: true, Optional: true}},
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia-modeset"), Device: true, Optional: true}},
-		// nvidia OpenCL/CUDA
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia-uvm"), Device: true, Optional: true}},
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia-uvm-tools"), Device: true, Optional: true}},
-
-		// flatpak commit d2dff2875bb3b7e2cd92d8204088d743fd07f3ff
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia0"), Device: true, Optional: true}}, {FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia1"), Device: true, Optional: true}},
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia2"), Device: true, Optional: true}}, {FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia3"), Device: true, Optional: true}},
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia4"), Device: true, Optional: true}}, {FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia5"), Device: true, Optional: true}},
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia6"), Device: true, Optional: true}}, {FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia7"), Device: true, Optional: true}},
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia8"), Device: true, Optional: true}}, {FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia9"), Device: true, Optional: true}},
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia10"), Device: true, Optional: true}}, {FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia11"), Device: true, Optional: true}},
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia12"), Device: true, Optional: true}}, {FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia13"), Device: true, Optional: true}},
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia14"), Device: true, Optional: true}}, {FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia15"), Device: true, Optional: true}},
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia16"), Device: true, Optional: true}}, {FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia17"), Device: true, Optional: true}},
-		{FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia18"), Device: true, Optional: true}}, {FilesystemConfig: &hst.FSBind{Source: container.AbsFHSDev.Append("nvidia19"), Device: true, Optional: true}},
-	}...)
-}

cmd/hpkg/proc.go

@@ -1,60 +0,0 @@
-package main
-
-import (
-	"context"
-	"encoding/json"
-	"errors"
-	"io"
-	"log"
-	"os"
-	"os/exec"
-
-	"hakurei.app/hst"
-	"hakurei.app/internal"
-	"hakurei.app/internal/hlog"
-)
-
-var hakureiPath = internal.MustHakureiPath()
-
-func mustRunApp(ctx context.Context, config *hst.Config, beforeFail func()) {
-	var (
-		cmd *exec.Cmd
-		st  io.WriteCloser
-	)
-
-	if r, w, err := os.Pipe(); err != nil {
-		beforeFail()
-		log.Fatalf("cannot pipe: %v", err)
-	} else {
-		if hlog.Load() {
-			cmd = exec.CommandContext(ctx, hakureiPath, "-v", "app", "3")
-		} else {
-			cmd = exec.CommandContext(ctx, hakureiPath, "app", "3")
-		}
-		cmd.Stdin, cmd.Stdout, cmd.Stderr = os.Stdin, os.Stdout, os.Stderr
-		cmd.ExtraFiles = []*os.File{r}
-		st = w
-	}
-
-	go func() {
-		if err := json.NewEncoder(st).Encode(config); err != nil {
-			beforeFail()
-			log.Fatalf("cannot send configuration: %v", err)
-		}
-	}()
-
-	if err := cmd.Start(); err != nil {
-		beforeFail()
-		log.Fatalf("cannot start hakurei: %v", err)
-	}
-	if err := cmd.Wait(); err != nil {
-		var exitError *exec.ExitError
-		if errors.As(err, &exitError) {
-			beforeFail()
-			internal.Exit(exitError.ExitCode())
-		} else {
-			beforeFail()
-			log.Fatalf("cannot wait: %v", err)
-		}
-	}
-}

cmd/hpkg/test/configuration.nix

@@ -1,62 +0,0 @@
-{ pkgs, ... }:
-{
-  users.users = {
-    alice = {
-      isNormalUser = true;
-      description = "Alice Foobar";
-      password = "foobar";
-      uid = 1000;
-    };
-  };
-
-  home-manager.users.alice.home.stateVersion = "24.11";
-
-  # Automatically login on tty1 as a normal user:
-  services.getty.autologinUser = "alice";
-
-  environment = {
-    variables = {
-      SWAYSOCK = "/tmp/sway-ipc.sock";
-      WLR_RENDERER = "pixman";
-    };
-  };
-
-  # Automatically configure and start Sway when logging in on tty1:
-  programs.bash.loginShellInit = ''
-    if [ "$(tty)" = "/dev/tty1" ]; then
-      set -e
-
-      mkdir -p ~/.config/sway
-      (sed s/Mod4/Mod1/ /etc/sway/config &&
-      echo 'output * bg ${pkgs.nixos-artwork.wallpapers.simple-light-gray.gnomeFilePath} fill' &&
-      echo 'output Virtual-1 res 1680x1050') > ~/.config/sway/config
-
-      sway --validate
-      systemd-cat --identifier=session sway && touch /tmp/sway-exit-ok
-    fi
-  '';
-
-  programs.sway.enable = true;
-
-  virtualisation = {
-    diskSize = 6 * 1024;
-
-    qemu.options = [
-      # Need to switch to a different GPU driver than the default one (-vga std) so that Sway can launch:
-      "-vga none -device virtio-gpu-pci"
-
-      # Increase zstd performance:
-      "-smp 8"
-    ];
-  };
-
-  environment.hakurei = {
-    enable = true;
-    stateDir = "/var/lib/hakurei";
-    users.alice = 0;
-
-    extraHomeConfig = {
-      home.stateVersion = "23.05";
-    };
-  };
-}

cmd/hpkg/test/default.nix

@@ -1,34 +0,0 @@
-{
-  nixosTest,
-  callPackage,
-
-  system,
-  self,
-}:
-let
-  buildPackage = self.buildPackage.${system};
-in
-nixosTest {
-  name = "hpkg";
-  nodes.machine = {
-    environment.etc = {
-      "foot.pkg".source = callPackage ./foot.nix { inherit buildPackage; };
-    };
-
-    imports = [
-      ./configuration.nix
-
-      self.nixosModules.hakurei
-      self.inputs.home-manager.nixosModules.home-manager
-    ];
-  };
-
-  # adapted from nixos sway integration tests
-
-  # testScriptWithTypes:49: error: Cannot call function of unknown type
-  #           (machine.succeed if succeed else machine.execute)(
-  #           ^
-  # Found 1 error in 1 file (checked 1 source file)
-  skipTypeCheck = true;
-  testScript = builtins.readFile ./test.py;
-}

cmd/hpkg/test/foot.nix

@@ -1,48 +0,0 @@
-{
-  lib,
-  buildPackage,
-  foot,
-  wayland-utils,
-  inconsolata,
-}:
-
-buildPackage {
-  name = "foot";
-  inherit (foot) version;
-
-  identity = 2;
-  id = "org.codeberg.dnkl.foot";
-
-  modules = [
-    {
-      home.packages = [
-        foot
-
-        # For wayland-info:
-        wayland-utils
-      ];
-    }
-  ];
-
-  nixosModules = [
-    {
-      # To help with OCR:
-      environment.etc."xdg/foot/foot.ini".text = lib.generators.toINI { } {
-        main = {
-          font = "inconsolata:size=14";
-        };
-        colors = rec {
-          foreground = "000000";
-          background = "ffffff";
-          regular2 = foreground;
-        };
-      };
-
-      fonts.packages = [ inconsolata ];
-    }
-  ];
-
-  script = ''
-    exec foot "$@"
-  '';
-}

cmd/hpkg/test/test.py

@@ -1,108 +0,0 @@
-import json
-import shlex
-
-q = shlex.quote
-NODE_GROUPS = ["nodes", "floating_nodes"]
-
-
-def swaymsg(command: str = "", succeed=True, type="command"):
-    assert command != "" or type != "command", "Must specify command or type"
-    shell = q(f"swaymsg -t {q(type)} -- {q(command)}")
-    with machine.nested(
-            f"sending swaymsg {shell!r}" + " (allowed to fail)" * (not succeed)
-    ):
-        ret = (machine.succeed if succeed else machine.execute)(
-            f"su - alice -c {shell}"
-        )
-
-    # execute also returns a status code, but disregard.
-    if not succeed:
-        _, ret = ret
-
-    if not succeed and not ret:
-        return None
-
-    parsed = json.loads(ret)
-    return parsed
-
-
-def walk(tree):
-    yield tree
-    for group in NODE_GROUPS:
-        for node in tree.get(group, []):
-            yield from walk(node)
-
-
-def wait_for_window(pattern):
-    def func(last_chance):
-        nodes = (node["name"] for node in walk(swaymsg(type="get_tree")))
-
-        if last_chance:
-            nodes = list(nodes)
-            machine.log(f"Last call! Current list of windows: {nodes}")
-
-        return any(pattern in name for name in nodes)
-
-    retry(func)
-
-
-def collect_state_ui(name):
-    swaymsg(f"exec hakurei ps > '/tmp/{name}.ps'")
-    machine.copy_from_vm(f"/tmp/{name}.ps", "")
-    swaymsg(f"exec hakurei --json ps > '/tmp/{name}.json'")
-    machine.copy_from_vm(f"/tmp/{name}.json", "")
-    machine.screenshot(name)
-
-
-def check_state(name, enablements):
-    instances = json.loads(machine.succeed("sudo -u alice -i XDG_RUNTIME_DIR=/run/user/1000 hakurei --json ps"))
-    if len(instances) != 1:
-        raise Exception(f"unexpected state length {len(instances)}")
-    instance = next(iter(instances.values()))
-
-    config = instance['config']
-
-    if len(config['args']) != 1 or not (config['args'][0].startswith("/nix/store/")) or f"hakurei-{name}-" not in (config['args'][0]):
-        raise Exception(f"unexpected args {instance['config']['args']}")
-
-    if config['enablements'] != enablements:
-        raise Exception(f"unexpected enablements {instance['config']['enablements']}")
-
-
-start_all()
-machine.wait_for_unit("multi-user.target")
-
-# To check hakurei's version:
-print(machine.succeed("sudo -u alice -i hakurei version"))
-
-# Wait for Sway to complete startup:
-machine.wait_for_file("/run/user/1000/wayland-1")
-machine.wait_for_file("/tmp/sway-ipc.sock")
-
-# Prepare hpkg directory:
-machine.succeed("install -dm 0700 -o alice -g users /var/lib/hakurei/1000")
-
-# Install hpkg app:
-swaymsg("exec hpkg -v install /etc/foot.pkg && touch /tmp/hpkg-install-ok")
-machine.wait_for_file("/tmp/hpkg-install-ok")
-
-# Start app (foot) with Wayland enablement:
-swaymsg("exec hpkg -v start org.codeberg.dnkl.foot")
-wait_for_window("hakurei@machine-foot")
-machine.send_chars("clear; wayland-info && touch /tmp/success-client\n")
-machine.wait_for_file("/tmp/hakurei.0/tmpdir/2/success-client")
-collect_state_ui("app_wayland")
-check_state("foot", {"wayland": True, "dbus": True, "pulse": True})
-# Verify acl on XDG_RUNTIME_DIR:
-print(machine.succeed("getfacl --absolute-names --omit-header --numeric /run/user/1000 | grep 1000002"))
-machine.send_chars("exit\n")
-machine.wait_until_fails("pgrep foot")
-# Verify acl cleanup on XDG_RUNTIME_DIR:
-machine.wait_until_fails("getfacl --absolute-names --omit-header --numeric /run/user/1000 | grep 1000002")
-
-# Exit Sway and verify process exit status 0:
-swaymsg("exit", succeed=False)
-machine.wait_for_file("/tmp/sway-exit-ok")
-
-# Print hakurei runDir contents:
-print(machine.succeed("find /run/user/1000/hakurei"))

cmd/hpkg/with.go

@@ -1,108 +0,0 @@
-package main
-
-import (
-	"context"
-	"strings"
-
-	"hakurei.app/container"
-	"hakurei.app/container/seccomp"
-	"hakurei.app/hst"
-	"hakurei.app/internal"
-)
-
-func withNixDaemon(
-	ctx context.Context,
-	action string, command []string, net bool, updateConfig func(config *hst.Config) *hst.Config,
-	app *appInfo, pathSet *appPathSet, dropShell bool, beforeFail func(),
-) {
-	mustRunAppDropShell(ctx, updateConfig(&hst.Config{
-		ID: app.ID,
-
-		Path: pathShell,
-		Args: []string{bash, "-lc", "rm -f /nix/var/nix/daemon-socket/socket && " +
-			// start nix-daemon
-			"nix-daemon --store / & " +
-			// wait for socket to appear
-			"(while [ ! -S /nix/var/nix/daemon-socket/socket ]; do sleep 0.01; done) && " +
-			// create directory so nix stops complaining
-			"mkdir -p /nix/var/nix/profiles/per-user/root/channels && " +
-			strings.Join(command, " && ") +
-			// terminate nix-daemon
-			" && pkill nix-daemon",
-		},
-
-		Username: "hakurei",
-		Shell:    pathShell,
-		Home:     pathDataData.Append(app.ID),
-		ExtraPerms: []*hst.ExtraPermConfig{
-			{Path: dataHome, Execute: true},
-			{Ensure: true, Path: pathSet.baseDir, Read: true, Write: true, Execute: true},
-		},
-
-		Identity: app.Identity,
-
-		Container: &hst.ContainerConfig{
-			Hostname:     formatHostname(app.Name) + "-" + action,
-			Userns:       true, // nix sandbox requires userns
-			HostNet:      net,
-			SeccompFlags: seccomp.AllowMultiarch,
-			Tty:          dropShell,
-			Filesystem: []hst.FilesystemConfigJSON{
-				{FilesystemConfig: &hst.FSBind{Target: container.AbsFHSEtc, Source: pathSet.cacheDir.Append("etc"), Special: true}},
-				{FilesystemConfig: &hst.FSBind{Source: pathSet.nixPath, Target: pathNix, Write: true}},
-				{FilesystemConfig: &hst.FSLink{Target: pathCurrentSystem, Linkname: app.CurrentSystem.String()}},
-				{FilesystemConfig: &hst.FSLink{Target: pathBin, Linkname: pathSwBin.String()}},
-				{FilesystemConfig: &hst.FSLink{Target: container.AbsFHSUsrBin, Linkname: pathSwBin.String()}},
-				{FilesystemConfig: &hst.FSBind{Target: pathDataData.Append(app.ID), Source: pathSet.homeDir, Write: true, Ensure: true}},
-			},
-		},
-	}), dropShell, beforeFail)
-}
-
-func withCacheDir(
-	ctx context.Context,
-	action string, command []string, workDir *container.Absolute,
-	app *appInfo, pathSet *appPathSet, dropShell bool, beforeFail func()) {
-	mustRunAppDropShell(ctx, &hst.Config{
-		ID: app.ID,
-
-		Path: pathShell,
-		Args: []string{bash, "-lc", strings.Join(command, " && ")},
-
-		Username: "nixos",
-		Shell:    pathShell,
-		Home:     pathDataData.Append(app.ID, "cache"),
-		ExtraPerms: []*hst.ExtraPermConfig{
-			{Path: dataHome, Execute: true},
-			{Ensure: true, Path: pathSet.baseDir, Read: true, Write: true, Execute: true},
-			{Path: workDir, Execute: true},
-		},
-
-		Identity: app.Identity,
-
-		Container: &hst.ContainerConfig{
-			Hostname:     formatHostname(app.Name) + "-" + action,
-			SeccompFlags: seccomp.AllowMultiarch,
-			Tty:          dropShell,
-			Filesystem: []hst.FilesystemConfigJSON{
-				{FilesystemConfig: &hst.FSBind{Target: container.AbsFHSEtc, Source: workDir.Append(container.FHSEtc), Special: true}},
-				{FilesystemConfig: &hst.FSBind{Source: workDir.Append("nix"), Target: pathNix}},
-				{FilesystemConfig: &hst.FSLink{Target: pathCurrentSystem, Linkname: app.CurrentSystem.String()}},
-				{FilesystemConfig: &hst.FSLink{Target: pathBin, Linkname: pathSwBin.String()}},
-				{FilesystemConfig: &hst.FSLink{Target: container.AbsFHSUsrBin, Linkname: pathSwBin.String()}},
-				{FilesystemConfig: &hst.FSBind{Source: workDir, Target: hst.AbsTmp.Append("bundle")}},
-				{FilesystemConfig: &hst.FSBind{Target: pathDataData.Append(app.ID, "cache"), Source: pathSet.cacheDir, Write: true, Ensure: true}},
-			},
-		},
-	}, dropShell, beforeFail)
-}
-
-func mustRunAppDropShell(ctx context.Context, config *hst.Config, dropShell bool, beforeFail func()) {
-	if dropShell {
-		config.Args = []string{bash, "-l"}
-		mustRunApp(ctx, config, beforeFail)
-		beforeFail()
-		internal.Exit(0)
-	}
-	mustRunApp(ctx, config, beforeFail)
-}

cmd/hsu/conf.go

@@ -0,0 +1,7 @@
+//go:build !rosa
+
+package main
+
+// hsuConfPath is an absolute pathname to the hsu configuration file. Its
+// contents are interpreted by parseConfig.
+const hsuConfPath = "/etc/hsurc"

cmd/hsu/config_rosa.go

@@ -0,0 +1,7 @@
+//go:build rosa
+
+package main
+
+// hsuConfPath is the pathname to the hsu configuration file, specific to
+// Rosa OS. Its contents are interpreted by parseConfig.
+const hsuConfPath = "/system/etc/hsurc"

cmd/hsu/hst.go

@@ -0,0 +1,16 @@
+package main
+
+/* keep in sync with hst */
+
+const (
+	userOffset = 100000
+	rangeSize  = userOffset / 10
+
+	identityStart = 0
+	identityEnd   = appEnd - appStart
+
+	appStart = rangeSize * 1
+	appEnd   = appStart + rangeSize - 1
+)
+
+func toUser(userid, appid uint32) uint32 { return userid*userOffset + appStart + appid }

cmd/hsu/main.go

@@ -1,3 +1,47 @@
+// hsu starts the hakurei shim as the target subordinate user.
+//
+// The hsu program must be installed with the setuid and setgid bit set, and
+// owned by root. A configuration file must be installed at /etc/hsurc with
+// permission bits 0400, and owned by root. Each line of the file specifies a
+// hakurei userid to kernel uid mapping. A line consists of the decimal string
+// representation of the uid of the user wishing to start hakurei containers,
+// followed by a space, followed by the decimal string representation of its
+// userid. Duplicate uid entries are ignored, with the first occurrence taking
+// effect.
+//
+// For example, to map the kernel uid 1000 to the hakurei user id 0:
+//
+//	1000 0
+//
+// # Internals
+//
+// Hakurei and hsu holds pathnames pointing to each other set at link time. For
+// this reason, a distribution of hakurei has fixed installation prefix. Since
+// this program is never invoked by the user, behaviour described in the
+// following paragraphs are considered an internal detail and not covered by the
+// compatibility promise.
+//
+// Since target container environment variables are set up in shim via the
+// [container] infrastructure, the environment is used for parameters from the
+// parent process.
+//
+// HAKUREI_SHIM specifies a single byte between '3' and '9' representing the
+// setup pipe file descriptor. It is passed as is to the shim process and is the
+// only value in the environment of the shim process. Since hsurc is not
+// accessible to the parent process, leaving this unset causes hsu to print the
+// corresponding hakurei user id of the parent and terminate.
+//
+// HAKUREI_IDENTITY specifies the identity of the instance being started and is
+// used to produce the kernel uid alongside hakurei user id looked up from hsurc.
+//
+// HAKUREI_GROUPS specifies supplementary groups to inherit from the credentials
+// of the parent process in a ' ' separated list of decimal string
+// representations of gid. This has the unfortunate consequence of allowing
+// users mapped via hsurc to effectively drop group membership, so special care
+// must be taken to ensure this does not lead to an increase in access. This is
+// not applicable to Rosa OS since unsigned code execution is not permitted
+// outside hakurei containers, and is generally nonapplicable to the security
+// model of hakurei, where all untrusted code runs within containers.
 package main
 
 import (
@@ -5,55 +49,56 @@ import (
 	"fmt"
 	"log"
 	"os"
-	"path"
+	"path/filepath"
+	"runtime"
 	"slices"
 	"strconv"
-	"strings"
 	"syscall"
 )
 
 const (
-	hsuConfFile = "/etc/hsurc"
-	envShim     = "HAKUREI_SHIM"
-	envAID      = "HAKUREI_APP_ID"
-	envGroups   = "HAKUREI_GROUPS"
-
-	PR_SET_NO_NEW_PRIVS = 0x26
+	// envShim is the name of the environment variable holding a single byte
+	// representing the shim setup pipe file descriptor.
+	envShim = "HAKUREI_SHIM"
+	// envIdentity is the name of the environment variable holding a decimal
+	// string representation of the current application identity.
+	envIdentity = "HAKUREI_IDENTITY"
+	// envGroups holds a ' ' separated list of decimal string representations of
+	// supplementary group gid. Membership requirements are enforced.
+	envGroups = "HAKUREI_GROUPS"
 )
 
+// hakureiPath is the absolute path to Hakurei.
+//
+// This is set by the linker.
+var hakureiPath string
+
 func main() {
+	const PR_SET_NO_NEW_PRIVS = 0x26
+	runtime.LockOSThread()
+
 	log.SetFlags(0)
 	log.SetPrefix("hsu: ")
-	log.SetOutput(os.Stderr)
 
 	if os.Geteuid() != 0 {
 		log.Fatal("this program must be owned by uid 0 and have the setuid bit set")
 	}
+	if os.Getegid() != os.Getgid() {
+		log.Fatal("this program must not have the setgid bit set")
+	}
 
 	puid := os.Getuid()
 	if puid == 0 {
 		log.Fatal("this program must not be started by root")
 	}
 
-	var toolPath string
-	pexe := path.Join("/proc", strconv.Itoa(os.Getppid()), "exe")
-	if p, err := os.Readlink(pexe); err != nil {
-		log.Fatalf("cannot read parent executable path: %v", err)
-	} else if strings.HasSuffix(p, " (deleted)") {
-		log.Fatal("hakurei executable has been deleted")
-	} else if p != mustCheckPath(hmain) {
-		log.Fatal("this program must be started by hakurei")
-	} else {
-		toolPath = p
+	if !filepath.IsAbs(hakureiPath) {
+		log.Fatal("this program is compiled incorrectly")
+		return
 	}
 
-	// uid = 1000000 +
-	//   fid * 10000 +
-	//           aid
-	uid := 1000000
-
 	// refuse to run if hsurc is not protected correctly
-	if s, err := os.Stat(hsuConfFile); err != nil {
+	if s, err := os.Stat(hsuConfPath); err != nil {
 		log.Fatal(err)
 	} else if s.Mode().Perm() != 0400 {
 		log.Fatal("bad hsurc perm")
@@ -62,29 +107,13 @@ func main() {
 	}
 
 	// authenticate before accepting user input
-	if f, err := os.Open(hsuConfFile); err != nil {
-		log.Fatal(err)
-	} else if fid, ok := mustParseConfig(f, puid); !ok {
-		log.Fatalf("uid %d is not in the hsurc file", puid)
-	} else {
-		uid += fid * 10000
-	}
-
-	// allowed aid range 0 to 9999
-	if as, ok := os.LookupEnv(envAID); !ok {
-		log.Fatal("HAKUREI_APP_ID not set")
-	} else if aid, err := parseUint32Fast(as); err != nil || aid < 0 || aid > 9999 {
-		log.Fatal("invalid aid")
-	} else {
-		uid += aid
-	}
+	userid := mustParseConfig(puid)
 
 	// pass through setup fd to shim
 	var shimSetupFd string
 	if s, ok := os.LookupEnv(envShim); !ok {
-		// hakurei requests target uid
-		// print resolved uid and exit
-		fmt.Print(uid)
+		// hakurei requests hsurc user id
+		fmt.Print(userid)
 		os.Exit(0)
 	} else if len(s) != 1 || s[0] > '9' || s[0] < '3' {
 		log.Fatal("HAKUREI_SHIM holds an invalid value")
@@ -92,6 +121,22 @@ func main() {
 		shimSetupFd = s
 	}
 
+	// start is going ahead at this point
+	identity := mustReadIdentity()
+
+	const (
+		// first possible uid outcome
+		uidStart = 10000
+		// last possible uid outcome
+		uidEnd = 999919999
+	)
+	uid := int(toUser(userid, identity))
+
+	// final bounds check to catch any bugs
+	if uid < uidStart || uid >= uidEnd {
+		panic("uid out of bounds")
+	}
+
 	// supplementary groups
 	var suppGroups, suppCurrent []int
 
@@ -119,13 +164,7 @@ func main() {
 		suppGroups = []int{uid}
 	}
 
-	// final bounds check to catch any bugs
-	if uid < 1000000 || uid >= 2000000 {
-		panic("uid out of bounds")
-	}
-
 	// careful! users in the allowlist is effectively allowed to drop groups via hsu
-
 	if err := syscall.Setresgid(uid, uid, uid); err != nil {
 		log.Fatalf("cannot set gid: %v", err)
 	}
@@ -135,10 +174,21 @@ func main() {
 	if err := syscall.Setresuid(uid, uid, uid); err != nil {
 		log.Fatalf("cannot set uid: %v", err)
 	}
-	if _, _, errno := syscall.AllThreadsSyscall(syscall.SYS_PRCTL, PR_SET_NO_NEW_PRIVS, 1, 0); errno != 0 {
+
+	if _, _, errno := syscall.AllThreadsSyscall(
+		syscall.SYS_PRCTL,
+		PR_SET_NO_NEW_PRIVS, 1,
+		0,
+	); errno != 0 {
 		log.Fatalf("cannot set no_new_privs flag: %s", errno.Error())
 	}
-	if err := syscall.Exec(toolPath, []string{"hakurei", "shim"}, []string{envShim + "=" + shimSetupFd}); err != nil {
+
+	if err := syscall.Exec(hakureiPath, []string{
+		"hakurei",
+		"shim",
+	}, []string{
+		envShim + "=" + shimSetupFd,
+	}); err != nil {
 		log.Fatalf("cannot start shim: %v", err)
 	}
 

cmd/hsu/package.nix

@@ -19,5 +19,5 @@ buildGoModule {
   ldflags = lib.attrsets.foldlAttrs (
     ldflags: name: value:
     ldflags ++ [ "-X main.${name}=${value}" ]
-  ) [ "-s -w" ] { hmain = "${hakurei}/libexec/hakurei"; };
+  ) [ "-s -w" ] { hakureiPath = "${hakurei}/libexec/hakurei"; };
 }

cmd/hsu/parse.go

@@ -6,62 +6,123 @@ import (
 	"fmt"
 	"io"
 	"log"
+	"math"
+	"os"
 	"strings"
 )
 
-func parseUint32Fast(s string) (int, error) {
+const (
+	// useridStart is the first userid.
+	useridStart = 0
+	// useridEnd is the last userid.
+	useridEnd = useridStart + rangeSize - 1
+)
+
+// parseUint32Fast parses a string representation of an unsigned 32-bit integer
+// value using the fast path only. This limits the range of values it is defined
+// in but is perfectly adequate for this use case.
+func parseUint32Fast(s string) (uint32, error) {
 	sLen := len(s)
 	if sLen < 1 {
-		return -1, errors.New("zero length string")
+		return 0, errors.New("zero length string")
 	}
 	if sLen > 10 {
-		return -1, errors.New("string too long")
+		return 0, errors.New("string too long")
 	}
 
-	n := 0
+	var n uint32
 	for i, ch := range []byte(s) {
 		ch -= '0'
 		if ch > 9 {
-			return -1, fmt.Errorf("invalid character '%s' at index %d", string(ch+'0'), i)
+			return 0, fmt.Errorf("invalid character '%s' at index %d", string(ch+'0'), i)
 		}
-		n = n*10 + int(ch)
+		n = n*10 + uint32(ch)
 	}
 	return n, nil
 }
 
-func parseConfig(r io.Reader, puid int) (fid int, ok bool, err error) {
+// parseConfig reads a list of allowed users from r until it encounters puid or
+// [io.EOF].
+//
+// Each line of the file specifies a hakurei userid to kernel uid mapping. A
+// line consists of the string representation of the uid of the user wishing to
+// start hakurei containers, followed by a space, followed by the string
+// representation of its userid. Duplicate uid entries are ignored, with the
+// first occurrence taking effect.
+//
+// All string representations are parsed by calling parseUint32Fast.
+func parseConfig(r io.Reader, puid uint32) (userid uint32, ok bool, err error) {
 	s := bufio.NewScanner(r)
-	var line, puid0 int
+	var (
+		line  uintptr
+		puid0 uint32
+	)
 	for s.Scan() {
 		line++
 
-		// <puid> <fid>
+		// <puid> <userid>
 		lf := strings.SplitN(s.Text(), " ", 2)
 		if len(lf) != 2 {
-			return -1, false, fmt.Errorf("invalid entry on line %d", line)
+			return useridEnd + 1, false, fmt.Errorf("invalid entry on line %d", line)
 		}
 
 		puid0, err = parseUint32Fast(lf[0])
 		if err != nil || puid0 < 1 {
-			return -1, false, fmt.Errorf("invalid parent uid on line %d", line)
+			return useridEnd + 1, false, fmt.Errorf("invalid parent uid on line %d", line)
 		}
 
 		ok = puid0 == puid
 		if ok {
-			// allowed fid range 0 to 99
-			if fid, err = parseUint32Fast(lf[1]); err != nil || fid < 0 || fid > 99 {
-				return -1, false, fmt.Errorf("invalid identity on line %d", line)
+			// userid bound to a range, uint32 size allows this to be increased if needed
+			if userid, err = parseUint32Fast(lf[1]); err != nil ||
+				userid < useridStart || userid > useridEnd {
+				return useridEnd + 1, false, fmt.Errorf("invalid userid on line %d", line)
 			}
 			return
 		}
 	}
-	return -1, false, s.Err()
+	return useridEnd + 1, false, s.Err()
 }
 
-func mustParseConfig(r io.Reader, puid int) (int, bool) {
-	fid, ok, err := parseConfig(r, puid)
-	if err != nil {
+// mustParseConfig calls parseConfig to interpret the contents of hsuConfPath,
+// terminating the program if an error is encountered, the syntax is incorrect,
+// or the current user is not authorised to use hsu because its uid is missing.
+//
+// Therefore, code after this function call can assume an authenticated state.
+//
+// mustParseConfig returns the userid value of the current user.
+func mustParseConfig(puid int) (userid uint32) {
+	if puid > math.MaxUint32 {
+		log.Fatalf("got impossible uid %d", puid)
+	}
+
+	var ok bool
+	if f, err := os.Open(hsuConfPath); err != nil {
+		log.Fatal(err)
+	} else if userid, ok, err = parseConfig(f, uint32(puid)); err != nil {
+		log.Fatal(err)
+	} else if err = f.Close(); err != nil {
 		log.Fatal(err)
 	}
-	return fid, ok
+	if !ok {
+		log.Fatalf("uid %d is not in the hsurc file", puid)
+	}
+
+	return
+}
+
+// mustReadIdentity calls parseUint32Fast to interpret the value stored in envIdentity,
+// terminating the program if the value is not set, malformed, or out of bounds.
+func mustReadIdentity() uint32 {
+	// ranges defined in hst and copied to this package to avoid importing hst
+	if as, ok := os.LookupEnv(envIdentity); !ok {
+		log.Fatal("HAKUREI_IDENTITY not set")
+		panic("unreachable")
+	} else if identity, err := parseUint32Fast(as); err != nil ||
+		identity < identityStart || identity > identityEnd {
+		log.Fatal("invalid identity")
+		panic("unreachable")
+	} else {
+		return identity
+	}
 }

cmd/hsu/parse_test.go

@@ -2,94 +2,105 @@ package main
 
 import (
 	"bytes"
+	"math"
 	"strconv"
 	"testing"
 )
 
-func Test_parseUint32Fast(t *testing.T) {
+func TestParseUint32Fast(t *testing.T) {
+	t.Parallel()
+
 	t.Run("zero-length", func(t *testing.T) {
+		t.Parallel()
+
 		if _, err := parseUint32Fast(""); err == nil || err.Error() != "zero length string" {
 			t.Errorf(`parseUint32Fast(""): error = %v`, err)
 			return
 		}
 	})
+
 	t.Run("overflow", func(t *testing.T) {
+		t.Parallel()
+
 		if _, err := parseUint32Fast("10000000000"); err == nil || err.Error() != "string too long" {
 			t.Errorf("parseUint32Fast: error = %v", err)
 			return
 		}
 	})
+
 	t.Run("invalid byte", func(t *testing.T) {
+		t.Parallel()
+
 		if _, err := parseUint32Fast("meow"); err == nil || err.Error() != "invalid character 'm' at index 0" {
 			t.Errorf(`parseUint32Fast("meow"): error = %v`, err)
 			return
 		}
 	})
-	t.Run("full range", func(t *testing.T) {
-		testRange := func(i, end int) {
+
+	t.Run("range", func(t *testing.T) {
+		t.Parallel()
+
+		testRange := func(i, end uint32) {
 			for ; i < end; i++ {
-				s := strconv.Itoa(i)
+				s := strconv.Itoa(int(i))
 				w := i
 				t.Run("parse "+s, func(t *testing.T) {
 					t.Parallel()
+
 					v, err := parseUint32Fast(s)
 					if err != nil {
-						t.Errorf("parseUint32Fast(%q): error = %v",
-							s, err)
+						t.Errorf("parseUint32Fast(%q): error = %v", s, err)
 						return
 					}
 					if v != w {
-						t.Errorf("parseUint32Fast(%q): got %v",
-							s, v)
+						t.Errorf("parseUint32Fast(%q): got %v", s, v)
 						return
 					}
 				})
 			}
 		}
 
-		testRange(0, 5000)
-		testRange(105000, 110000)
-		testRange(23005000, 23010000)
-		testRange(456005000, 456010000)
-		testRange(7890005000, 7890010000)
+		testRange(0, 2500)
+		testRange(23002500, 23005000)
+		testRange(math.MaxUint32-2500, math.MaxUint32)
 	})
 }
 
-func Test_parseConfig(t *testing.T) {
+func TestParseConfig(t *testing.T) {
+	t.Parallel()
+
 	testCases := []struct {
 		name       string
-		puid, want int
+		puid, want uint32
 		wantErr    string
 		rc         string
 	}{
-		{"empty", 0, -1, "", ``},
-		{"invalid field", 0, -1, "invalid entry on line 1", `9`},
-		{"invalid puid", 0, -1, "invalid parent uid on line 1", `f 9`},
-		{"invalid fid", 1000, -1, "invalid identity on line 1", `1000 f`},
+		{"empty", 0, useridEnd + 1, "", ``},
+		{"invalid field", 0, useridEnd + 1, "invalid entry on line 1", `9`},
+		{"invalid puid", 0, useridEnd + 1, "invalid parent uid on line 1", `f 9`},
+		{"invalid userid", 1000, useridEnd + 1, "invalid userid on line 1", `1000 f`},
 		{"match", 1000, 0, "", `1000 0`},
 	}
 
 	for _, tc := range testCases {
 		t.Run(tc.name, func(t *testing.T) {
-			fid, ok, err := parseConfig(bytes.NewBufferString(tc.rc), tc.puid)
+			t.Parallel()
+
+			userid, ok, err := parseConfig(bytes.NewBufferString(tc.rc), tc.puid)
 			if err == nil && tc.wantErr != "" {
-				t.Errorf("parseConfig: error = %v; wantErr %q",
-					err, tc.wantErr)
+				t.Errorf("parseConfig: error = %v; want %q", err, tc.wantErr)
 				return
 			}
 			if err != nil && err.Error() != tc.wantErr {
-				t.Errorf("parseConfig: error = %q; wantErr %q",
-					err, tc.wantErr)
+				t.Errorf("parseConfig: error = %q; want %q", err, tc.wantErr)
 				return
 			}
-			if ok == (tc.want == -1) {
-				t.Errorf("parseConfig: ok = %v; want %v",
-					ok, tc.want)
+			if ok == (tc.want == useridEnd+1) {
+				t.Errorf("parseConfig: ok = %v; want %v", ok, tc.want)
 				return
 			}
-			if fid != tc.want {
-				t.Errorf("parseConfig: fid = %v; want %v",
-					fid, tc.want)
+			if userid != tc.want {
+				t.Errorf("parseConfig: %v; want %v", userid, tc.want)
 			}
 		})
 	}

cmd/hsu/path.go

@@ -1,20 +0,0 @@
-package main
-
-import (
-	"log"
-	"path"
-)
-
-const compPoison = "INVALIDINVALIDINVALIDINVALIDINVALID"
-
-var (
-	hmain = compPoison
-)
-
-func mustCheckPath(p string) string {
-	if p != compPoison && p != "" && path.IsAbs(p) {
-		return p
-	}
-	log.Fatal("this program is compiled incorrectly")
-	return compPoison
-}

cmd/mbf/cache.go

@@ -0,0 +1,136 @@
+package main
+
+import (
+	"context"
+	"net/http"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"hakurei.app/check"
+	"hakurei.app/internal/pkg"
+	"hakurei.app/internal/rosa"
+	"hakurei.app/message"
+)
+
+// cache refers to an instance of [pkg.Cache] that might be open.
+type cache struct {
+	ctx context.Context
+	msg message.Msg
+
+	// Should generally not be used directly.
+	c *pkg.Cache
+
+	cures, jobs int
+	// Primarily to work around missing landlock LSM.
+	hostAbstract bool
+	// Set SCHED_IDLE.
+	idle bool
+	// Unset [pkg.CSuppressInit].
+	verboseInit bool
+	// Loaded artifact of [rosa.QEMU].
+	qemu pkg.Artifact
+
+	base, mirror string
+}
+
+// open opens the underlying [pkg.Cache].
+func (cache *cache) open() (err error) {
+	if cache.c != nil {
+		return os.ErrInvalid
+	}
+
+	var base *check.Absolute
+	if cache.base, err = filepath.Abs(cache.base); err != nil {
+		return
+	} else if base, err = check.NewAbs(cache.base); err != nil {
+		return
+	}
+
+	var flags int
+	if cache.idle {
+		flags |= pkg.CSchedIdle
+	}
+	if cache.hostAbstract {
+		flags |= pkg.CHostAbstract
+	}
+	if !cache.verboseInit {
+		flags |= pkg.CSuppressInit
+	}
+
+	done := make(chan struct{})
+	defer close(done)
+	go func() {
+		select {
+		case <-cache.ctx.Done():
+			if testing.Testing() {
+				return
+			}
+			os.Exit(2)
+
+		case <-done:
+			return
+		}
+	}()
+
+	cache.msg.Verbosef("opening cache at %s", base)
+	cache.c, err = pkg.Open(
+		cache.ctx,
+		cache.msg,
+		flags,
+		cache.cures,
+		cache.jobs,
+		base,
+	)
+	if err != nil {
+		return
+	}
+	done <- struct{}{}
+
+	if cache.mirror != "" {
+		var pub []byte
+		pub, err = os.ReadFile(base.Append("ed25519.pub").String())
+		if err != nil {
+			cache.c.Close()
+			return
+		}
+		var r rosa.Remote
+		if r, err = rosa.NewRemote(cache.mirror, pub, http.DefaultClient); err != nil {
+			cache.c.Close()
+			return err
+		}
+		cache.c.SetExternal(r)
+	}
+
+	if cache.qemu != nil {
+		var pathname *check.Absolute
+		pathname, _, err = cache.c.Cure(cache.qemu)
+		if err != nil {
+			cache.c.Close()
+			return
+		}
+
+		for arch, entry := range rosa.Arches(pathname) {
+			pkg.RegisterArch(arch, entry)
+		}
+	}
+
+	return
+}
+
+// Close closes the underlying [pkg.Cache] if it is open.
+func (cache *cache) Close() {
+	if cache.c != nil {
+		cache.c.Close()
+	}
+}
+
+// Do calls f on the underlying cache and returns its error value.
+func (cache *cache) Do(f func(cache *pkg.Cache) error) error {
+	if cache.c == nil {
+		if err := cache.open(); err != nil {
+			return err
+		}
+	}
+	return f(cache.c)
+}

cmd/mbf/cache_test.go

@@ -0,0 +1,37 @@
+package main
+
+import (
+	"log"
+	"os"
+	"testing"
+
+	"hakurei.app/internal/pkg"
+	"hakurei.app/message"
+)
+
+func TestCache(t *testing.T) {
+	t.Parallel()
+
+	cm := cache{
+		ctx:  t.Context(),
+		msg:  message.New(log.New(os.Stderr, "check: ", 0)),
+		base: t.TempDir(),
+
+		hostAbstract: true, idle: true,
+	}
+	defer cm.Close()
+	cm.Close()
+
+	if err := cm.open(); err != nil {
+		t.Fatalf("open: error = %v", err)
+	}
+	if err := cm.open(); err != os.ErrInvalid {
+		t.Errorf("(duplicate) open: error = %v", err)
+	}
+
+	if err := cm.Do(func(cache *pkg.Cache) error {
+		return cache.Scrub(0)
+	}); err != nil {
+		t.Errorf("Scrub: error = %v", err)
+	}
+}

cmd/mbf/daemon.go

@@ -0,0 +1,354 @@
+package main
+
+import (
+	"context"
+	"encoding/binary"
+	"errors"
+	"io"
+	"log"
+	"math"
+	"net"
+	"os"
+	"sync"
+	"syscall"
+	"testing"
+	"time"
+	"unique"
+
+	"hakurei.app/check"
+	"hakurei.app/internal/pkg"
+)
+
+// daemonTimeout is the maximum amount of time cureFromIR will wait on I/O.
+const daemonTimeout = 30 * time.Second
+
+// daemonDeadline returns the deadline corresponding to daemonTimeout, or the
+// zero value when running in a test.
+func daemonDeadline() time.Time {
+	if testing.Testing() {
+		return time.Time{}
+	}
+	return time.Now().Add(daemonTimeout)
+}
+
+const (
+	// remoteNoReply notifies that the client will not receive a cure reply.
+	remoteNoReply = 1 << iota
+)
+
+// cureFromIR services an IR curing request.
+func cureFromIR(
+	cache *pkg.Cache,
+	conn net.Conn,
+	flags uint64,
+) (pkg.Artifact, error) {
+	a, decodeErr := cache.NewDecoder(conn).Decode()
+	if decodeErr != nil {
+		_, err := conn.Write([]byte("\x00" + decodeErr.Error()))
+		return nil, errors.Join(decodeErr, err, conn.Close())
+	}
+
+	pathname, _, cureErr := cache.Cure(a)
+	if flags&remoteNoReply != 0 {
+		return a, errors.Join(cureErr, conn.Close())
+	}
+	if err := conn.SetWriteDeadline(daemonDeadline()); err != nil {
+		return a, errors.Join(cureErr, err, conn.Close())
+	}
+	if cureErr != nil {
+		_, err := conn.Write([]byte("\x00" + cureErr.Error()))
+		return a, errors.Join(cureErr, err, conn.Close())
+	}
+	_, err := conn.Write([]byte(pathname.String()))
+	if testing.Testing() && errors.Is(err, io.ErrClosedPipe) {
+		return a, nil
+	}
+	return a, errors.Join(err, conn.Close())
+}
+
+const (
+	// specialCancel is a message consisting of a single identifier referring
+	// to a curing artifact to be cancelled.
+	specialCancel = iota
+	// specialAbort requests for all pending cures to be aborted. It has no
+	// message body.
+	specialAbort
+
+	// remoteSpecial denotes a special message with custom layout.
+	remoteSpecial = math.MaxUint64
+)
+
+// writeSpecialHeader writes the header of a remoteSpecial message.
+func writeSpecialHeader(conn net.Conn, kind uint64) error {
+	var sh [16]byte
+	binary.LittleEndian.PutUint64(sh[:], remoteSpecial)
+	binary.LittleEndian.PutUint64(sh[8:], kind)
+	if n, err := conn.Write(sh[:]); err != nil {
+		return err
+	} else if n != len(sh) {
+		return io.ErrShortWrite
+	}
+	return nil
+}
+
+// cancelIdent reads an identifier from conn and cancels the corresponding cure.
+func cancelIdent(
+	cache *pkg.Cache,
+	conn net.Conn,
+) (*pkg.ID, bool, error) {
+	var ident pkg.ID
+	if _, err := io.ReadFull(conn, ident[:]); err != nil {
+		return nil, false, errors.Join(err, conn.Close())
+	}
+	ok := cache.Cancel(unique.Make(ident))
+	return &ident, ok, conn.Close()
+}
+
+// serve services connections from a [net.UnixListener].
+func serve(
+	ctx context.Context,
+	log *log.Logger,
+	cm *cache,
+	ul *net.UnixListener,
+) error {
+	ul.SetUnlinkOnClose(true)
+	if cm.c == nil {
+		if err := cm.open(); err != nil {
+			return errors.Join(err, ul.Close())
+		}
+	}
+
+	var wg sync.WaitGroup
+	defer wg.Wait()
+
+	wg.Go(func() {
+		for {
+			if ctx.Err() != nil {
+				break
+			}
+
+			conn, err := ul.AcceptUnix()
+			if err != nil {
+				if !errors.Is(err, os.ErrDeadlineExceeded) {
+					log.Println(err)
+				}
+				continue
+			}
+			wg.Go(func() {
+				done := make(chan struct{})
+				defer close(done)
+				go func() {
+					select {
+					case <-ctx.Done():
+						_ = conn.SetDeadline(time.Now())
+
+					case <-done:
+						return
+					}
+				}()
+
+				if _err := conn.SetReadDeadline(daemonDeadline()); _err != nil {
+					log.Println(_err)
+					if _err = conn.Close(); _err != nil {
+						log.Println(_err)
+					}
+					return
+				}
+
+				var word [8]byte
+				if _, _err := io.ReadFull(conn, word[:]); _err != nil {
+					log.Println(_err)
+					if _err = conn.Close(); _err != nil {
+						log.Println(_err)
+					}
+					return
+				}
+				flags := binary.LittleEndian.Uint64(word[:])
+
+				if flags == remoteSpecial {
+					if _, _err := io.ReadFull(conn, word[:]); _err != nil {
+						log.Println(_err)
+						if _err = conn.Close(); _err != nil {
+							log.Println(_err)
+						}
+						return
+					}
+					switch special := binary.LittleEndian.Uint64(word[:]); special {
+					default:
+						log.Printf("invalid special %d", special)
+
+					case specialCancel:
+						if id, ok, _err := cancelIdent(cm.c, conn); _err != nil {
+							log.Println(_err)
+						} else if !ok {
+							log.Println(
+								"attempting to cancel invalid artifact",
+								pkg.Encode(*id),
+							)
+						} else {
+							log.Println(
+								"cancelled artifact",
+								pkg.Encode(*id),
+							)
+						}
+
+					case specialAbort:
+						log.Println("aborting all pending cures")
+						cm.c.Abort()
+						if _err := conn.Close(); _err != nil {
+							log.Println(_err)
+						}
+					}
+
+					return
+				}
+
+				if a, _err := cureFromIR(cm.c, conn, flags); _err != nil {
+					log.Println(_err)
+				} else {
+					log.Printf(
+						"fulfilled artifact %s",
+						pkg.Encode(cm.c.Ident(a).Value()),
+					)
+				}
+			})
+		}
+	})
+
+	<-ctx.Done()
+	if err := ul.SetDeadline(time.Now()); err != nil {
+		return errors.Join(err, ul.Close())
+	}
+	wg.Wait()
+	return ul.Close()
+}
+
+// dial wraps [net.DialUnix] with a context.
+func dial(ctx context.Context, addr *net.UnixAddr) (
+	done chan<- struct{},
+	conn *net.UnixConn,
+	err error,
+) {
+	conn, err = net.DialUnix("unix", nil, addr)
+	if err != nil {
+		return
+	}
+
+	d := make(chan struct{})
+	done = d
+	go func() {
+		select {
+		case <-ctx.Done():
+			_ = conn.SetDeadline(time.Now())
+
+		case <-d:
+			return
+		}
+	}()
+	return
+}
+
+// cureRemote cures a [pkg.Artifact] on a daemon.
+func cureRemote(
+	ctx context.Context,
+	addr *net.UnixAddr,
+	a pkg.Artifact,
+	flags uint64,
+) (*check.Absolute, error) {
+	if flags == remoteSpecial {
+		return nil, syscall.EINVAL
+	}
+
+	done, conn, err := dial(ctx, addr)
+	if err != nil {
+		return nil, err
+	}
+	defer close(done)
+
+	if n, flagErr := conn.Write(binary.LittleEndian.AppendUint64(nil, flags)); flagErr != nil {
+		return nil, errors.Join(flagErr, conn.Close())
+	} else if n != 8 {
+		return nil, errors.Join(io.ErrShortWrite, conn.Close())
+	}
+
+	if err = pkg.NewIR().EncodeAll(conn, a); err != nil {
+		return nil, errors.Join(err, conn.Close())
+	} else if err = conn.CloseWrite(); err != nil {
+		return nil, errors.Join(err, conn.Close())
+	}
+
+	if flags&remoteNoReply != 0 {
+		return nil, conn.Close()
+	}
+
+	payload, recvErr := io.ReadAll(conn)
+	if err = errors.Join(recvErr, conn.Close()); err != nil {
+		if errors.Is(err, os.ErrDeadlineExceeded) {
+			if cancelErr := ctx.Err(); cancelErr != nil {
+				err = cancelErr
+			}
+		}
+		return nil, err
+	}
+
+	if len(payload) > 0 && payload[0] == 0 {
+		return nil, errors.New(string(payload[1:]))
+	}
+
+	var p *check.Absolute
+	p, err = check.NewAbs(string(payload))
+	return p, err
+}
+
+// cancelRemote cancels a [pkg.Artifact] curing on a daemon.
+func cancelRemote(
+	ctx context.Context,
+	addr *net.UnixAddr,
+	a pkg.Artifact,
+	wait bool,
+) error {
+	done, conn, err := dial(ctx, addr)
+	if err != nil {
+		return err
+	}
+	defer close(done)
+
+	if err = writeSpecialHeader(conn, specialCancel); err != nil {
+		return errors.Join(err, conn.Close())
+	}
+
+	var n int
+	id := pkg.NewIR().Ident(a).Value()
+	if n, err = conn.Write(id[:]); err != nil {
+		return errors.Join(err, conn.Close())
+	} else if n != len(id) {
+		return errors.Join(io.ErrShortWrite, conn.Close())
+	}
+	if wait {
+		if _, err = conn.Read(make([]byte, 1)); err == io.EOF {
+			err = nil
+		}
+	}
+	return errors.Join(err, conn.Close())
+}
+
+// abortRemote aborts all [pkg.Artifact] curing on a daemon.
+func abortRemote(
+	ctx context.Context,
+	addr *net.UnixAddr,
+	wait bool,
+) error {
+	done, conn, err := dial(ctx, addr)
+	if err != nil {
+		return err
+	}
+	defer close(done)
+
+	err = writeSpecialHeader(conn, specialAbort)
+	if wait && err == nil {
+		if _, err = conn.Read(make([]byte, 1)); err == io.EOF {
+			err = nil
+		}
+	}
+	return errors.Join(err, conn.Close())
+}

cmd/mbf/daemon_test.go

@@ -0,0 +1,146 @@
+package main
+
+import (
+	"bytes"
+	"context"
+	"errors"
+	"io"
+	"log"
+	"net"
+	"os"
+	"path/filepath"
+	"slices"
+	"strings"
+	"testing"
+	"time"
+
+	"hakurei.app/check"
+	"hakurei.app/internal/pkg"
+	"hakurei.app/message"
+)
+
+func TestNoReply(t *testing.T) {
+	t.Parallel()
+	if !daemonDeadline().IsZero() {
+		t.Fatal("daemonDeadline did not return the zero value")
+	}
+
+	c, err := pkg.Open(
+		t.Context(),
+		message.New(log.New(os.Stderr, "cir: ", 0)),
+		0, 0, 0,
+		check.MustAbs(t.TempDir()),
+	)
+	if err != nil {
+		t.Fatalf("Open: error = %v", err)
+	}
+	defer c.Close()
+
+	client, server := net.Pipe()
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		go func() {
+			<-t.Context().Done()
+			if _err := client.SetDeadline(time.Now()); _err != nil && !errors.Is(_err, io.ErrClosedPipe) {
+				panic(_err)
+			}
+		}()
+
+		if _err := c.EncodeAll(
+			client,
+			pkg.NewFile("check", []byte{0}),
+		); _err != nil {
+			panic(_err)
+		} else if _err = client.Close(); _err != nil {
+			panic(_err)
+		}
+	}()
+
+	a, cureErr := cureFromIR(c, server, remoteNoReply)
+	if cureErr != nil {
+		t.Fatalf("cureFromIR: error = %v", cureErr)
+	}
+
+	<-done
+	wantIdent := pkg.MustDecode("fiZf-ZY_Yq6qxJNrHbMiIPYCsGkUiKCRsZrcSELXTqZWtCnESlHmzV5ThhWWGGYG")
+	if gotIdent := c.Ident(a).Value(); gotIdent != wantIdent {
+		t.Errorf(
+			"cureFromIR: %s, want %s",
+			pkg.Encode(gotIdent), pkg.Encode(wantIdent),
+		)
+	}
+}
+
+func TestDaemon(t *testing.T) {
+	t.Parallel()
+
+	var buf bytes.Buffer
+	logger := log.New(&buf, "daemon: ", 0)
+
+	addr := net.UnixAddr{
+		Name: filepath.Join(t.TempDir(), "daemon"),
+		Net:  "unix",
+	}
+
+	ctx, cancel := context.WithCancel(t.Context())
+	defer cancel()
+
+	cm := cache{
+		ctx:  ctx,
+		msg:  message.New(logger),
+		base: t.TempDir(),
+	}
+	defer cm.Close()
+
+	ul, err := net.ListenUnix("unix", &addr)
+	if err != nil {
+		t.Fatalf("ListenUnix: error = %v", err)
+	}
+
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		if _err := serve(ctx, logger, &cm, ul); _err != nil {
+			panic(_err)
+		}
+	}()
+
+	if err = cancelRemote(ctx, &addr, pkg.NewFile("nonexistent", nil), true); err != nil {
+		t.Fatalf("cancelRemote: error = %v", err)
+	}
+
+	if err = abortRemote(ctx, &addr, true); err != nil {
+		t.Fatalf("abortRemote: error = %v", err)
+	}
+
+	// keep this last for synchronisation
+	var p *check.Absolute
+	p, err = cureRemote(ctx, &addr, pkg.NewFile("check", []byte{0}), 0)
+	if err != nil {
+		t.Fatalf("cureRemote: error = %v", err)
+	}
+
+	cancel()
+	<-done
+
+	const want = "fiZf-ZY_Yq6qxJNrHbMiIPYCsGkUiKCRsZrcSELXTqZWtCnESlHmzV5ThhWWGGYG"
+	if got := filepath.Base(p.String()); got != want {
+		t.Errorf("cureRemote: %s, want %s", got, want)
+	}
+
+	wantLog := []string{
+		"",
+		"daemon: aborting all pending cures",
+		"daemon: attempting to cancel invalid artifact kQm9fmnCmXST1-MMmxzcau2oKZCXXrlZydo4PkeV5hO_2PKfeC8t98hrbV_ZZx_j",
+		"daemon: fulfilled artifact fiZf-ZY_Yq6qxJNrHbMiIPYCsGkUiKCRsZrcSELXTqZWtCnESlHmzV5ThhWWGGYG",
+	}
+	gotLog := strings.Split(buf.String(), "\n")
+	slices.Sort(gotLog)
+	if !slices.Equal(gotLog, wantLog) {
+		t.Errorf(
+			"serve: logged\n%s\nwant\n%s",
+			strings.Join(gotLog, "\n"), strings.Join(wantLog, "\n"),
+		)
+	}
+}

cmd/mbf/info.go

@@ -0,0 +1,118 @@
+package main
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"os"
+	"strings"
+	"unique"
+
+	"hakurei.app/internal/pkg"
+	"hakurei.app/internal/rosa"
+)
+
+// commandInfo implements the info subcommand.
+func commandInfo(
+	cm *cache,
+	args []string,
+	w io.Writer,
+	writeStatus bool,
+	r *rosa.Report,
+) (err error) {
+	if len(args) == 0 {
+		return errors.New("info requires at least 1 argument")
+	}
+
+	// recovered by HandleAccess
+	mustPrintln := func(a ...any) {
+		if _, _err := fmt.Fprintln(w, a...); _err != nil {
+			panic(_err)
+		}
+	}
+	mustPrint := func(a ...any) {
+		if _, _err := fmt.Fprint(w, a...); _err != nil {
+			panic(_err)
+		}
+	}
+
+	t := rosa.Native().Std()
+	for i, name := range args {
+		handle := rosa.ArtifactH(unique.Make(name))
+		if meta, a := t.Load(handle); meta == nil {
+			return fmt.Errorf("unknown artifact %q", name)
+		} else {
+			var suffix string
+
+			if meta.Version != rosa.Unversioned {
+				suffix += "-" + meta.Version
+			}
+			mustPrintln("name        : " + name + suffix)
+
+			mustPrintln("description : " + meta.Description)
+			if meta.Website != "" {
+				mustPrintln("website     : " +
+					strings.TrimSuffix(meta.Website, "/"))
+			}
+			if len(meta.Dependencies) > 0 {
+				mustPrint("depends on  :")
+				for _, d := range meta.Dependencies {
+					_meta, _ := rosa.Native().Std().MustLoad(d)
+					s := _meta.Name
+					if _meta.Version != rosa.Unversioned {
+						s += "-" + _meta.Version
+					}
+					mustPrint(" " + s)
+				}
+				mustPrintln()
+			}
+
+			const statusPrefix = "status      : "
+			if writeStatus {
+				if r == nil {
+					var f io.ReadSeekCloser
+					err = cm.Do(func(cache *pkg.Cache) (err error) {
+						f, err = cache.OpenStatus(a)
+						return
+					})
+					if err != nil {
+						if errors.Is(err, os.ErrNotExist) {
+							mustPrintln(
+								statusPrefix + "not yet cured",
+							)
+						} else {
+							return
+						}
+					} else {
+						mustPrint(statusPrefix)
+						_, err = io.Copy(w, f)
+						if err = errors.Join(err, f.Close()); err != nil {
+							return
+						}
+					}
+				} else if err = cm.Do(func(cache *pkg.Cache) (err error) {
+					status, n := r.ArtifactOf(cache.Ident(a))
+					if status == nil {
+						mustPrintln(
+							statusPrefix + "not in report",
+						)
+					} else {
+						mustPrintln("size        :", n)
+						mustPrint(statusPrefix)
+						if _, err = w.Write(status); err != nil {
+							return
+						}
+					}
+					return
+				}); err != nil {
+					return
+				}
+			}
+
+			if i != len(args)-1 {
+				mustPrintln()
+			}
+		}
+	}
+	return nil
+}

cmd/mbf/info_test.go

@@ -0,0 +1,190 @@
+package main
+
+import (
+	"context"
+	"fmt"
+	"log"
+	"os"
+	"path/filepath"
+	"reflect"
+	"strings"
+	"syscall"
+	"testing"
+	"unique"
+	"unsafe"
+
+	"hakurei.app/internal/pkg"
+	"hakurei.app/internal/rosa"
+	"hakurei.app/message"
+)
+
+func TestInfo(t *testing.T) {
+	t.Parallel()
+
+	_t := rosa.Native().Std()
+	qemuMeta, _ := _t.Load(rosa.H("qemu"))
+	glibMeta, _ := _t.Load(rosa.H("glib"))
+	zlibMeta, zlib := _t.Load(rosa.H("zlib"))
+	zstdMeta, _ := _t.Load(rosa.H("zstd"))
+	hakureiMeta, _ := _t.Load(rosa.H("hakurei"))
+	hakureiDistMeta, _ := _t.Load(rosa.H("hakurei-dist"))
+
+	testCases := []struct {
+		name    string
+		args    []string
+		status  map[string]string
+		report  string
+		want    string
+		wantErr any
+	}{
+		{"qemu", []string{"qemu"}, nil, "", `
+name        : qemu-` + qemuMeta.Version + `
+description : a generic and open source machine emulator and virtualizer
+website     : https://www.qemu.org
+depends on  : glib-` + glibMeta.Version + ` zstd-` + zstdMeta.Version + `
+`, nil},
+
+		{"multi", []string{"hakurei", "hakurei-dist"}, nil, "", `
+name        : hakurei-` + hakureiMeta.Version + `
+description : low-level userspace tooling for Rosa OS
+website     : https://hakurei.app
+
+name        : hakurei-dist-` + hakureiDistMeta.Version + `
+description : low-level userspace tooling for Rosa OS (distribution tarball)
+website     : https://hakurei.app
+`, nil},
+
+		{"nonexistent", []string{"zlib", "\x00"}, nil, "", `
+name        : zlib-` + zlibMeta.Version + `
+description : lossless data-compression library
+website     : https://zlib.net
+
+`, fmt.Errorf("unknown artifact %q", "\x00")},
+
+		{"status cache", []string{"zlib", "zstd"}, map[string]string{
+			"zstd":    "internal/pkg (amd64) on satori\n",
+			"hakurei": "internal/pkg (amd64) on satori\n\n",
+		}, "", `
+name        : zlib-` + zlibMeta.Version + `
+description : lossless data-compression library
+website     : https://zlib.net
+status      : not yet cured
+
+name        : zstd-` + zstdMeta.Version + `
+description : a fast compression algorithm
+website     : https://facebook.github.io/zstd
+status      : internal/pkg (amd64) on satori
+`, nil},
+
+		{"status cache perm", []string{"zlib"}, map[string]string{
+			"zlib": "\x00",
+		}, "", `
+name        : zlib-` + zlibMeta.Version + `
+description : lossless data-compression library
+website     : https://zlib.net
+`, func(cm *cache) error {
+			return &os.PathError{
+				Op:   "open",
+				Path: filepath.Join(cm.base, "status", pkg.Encode(cm.c.Ident(zlib).Value())),
+				Err:  syscall.EACCES,
+			}
+		}},
+
+		{"status report", []string{"zlib"}, nil, strings.Repeat("\x00", len(pkg.Checksum{})+8), `
+name        : zlib-` + zlibMeta.Version + `
+description : lossless data-compression library
+website     : https://zlib.net
+status      : not in report
+`, nil},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			var (
+				cm  *cache
+				buf strings.Builder
+				r   *rosa.Report
+			)
+
+			if tc.status != nil || tc.report != "" {
+				cm = &cache{
+					ctx:  context.Background(),
+					msg:  message.New(log.New(os.Stderr, "info: ", 0)),
+					base: t.TempDir(),
+				}
+				defer cm.Close()
+			}
+
+			if tc.report != "" {
+				pathname := filepath.Join(t.TempDir(), "report")
+				err := os.WriteFile(
+					pathname,
+					unsafe.Slice(unsafe.StringData(tc.report), len(tc.report)),
+					0400,
+				)
+				if err != nil {
+					t.Fatal(err)
+				}
+
+				r, err = rosa.OpenReport(pathname)
+				if err != nil {
+					t.Fatal(err)
+				}
+				defer func() {
+					if err = r.Close(); err != nil {
+						t.Fatal(err)
+					}
+				}()
+			}
+
+			if tc.status != nil {
+				for name, status := range tc.status {
+					_, a := _t.Load(rosa.ArtifactH(unique.Make(name)))
+					if a == nil {
+						t.Fatalf("invalid name %q", name)
+					}
+					perm := os.FileMode(0400)
+					if status == "\x00" {
+						perm = 0
+					}
+					if err := cm.Do(func(cache *pkg.Cache) error {
+						return os.WriteFile(filepath.Join(
+							cm.base,
+							"status",
+							pkg.Encode(cache.Ident(a).Value()),
+						), unsafe.Slice(unsafe.StringData(status), len(status)), perm)
+					}); err != nil {
+						t.Fatalf("Do: error = %v", err)
+					}
+				}
+			}
+
+			var wantErr error
+			switch c := tc.wantErr.(type) {
+			case error:
+				wantErr = c
+			case func(cm *cache) error:
+				wantErr = c(cm)
+			default:
+				if tc.wantErr != nil {
+					t.Fatalf("invalid wantErr %#v", tc.wantErr)
+				}
+			}
+
+			if err := commandInfo(
+				cm,
+				tc.args,
+				&buf,
+				cm != nil,
+				r,
+			); !reflect.DeepEqual(err, wantErr) {
+				t.Fatalf("commandInfo: error = %v, want %v", err, wantErr)
+			}
+
+			if got := buf.String(); got != strings.TrimPrefix(tc.want, "\n") {
+				t.Errorf("commandInfo:\n%s\nwant\n%s", got, tc.want)
+			}
+		})
+	}
+}

cmd/mbf/internal/pkgserver/api.go

@@ -0,0 +1,202 @@
+// Package pkgserver implements the package metadata service backend.
+package pkgserver
+
+import (
+	"context"
+	"encoding/json"
+	"log"
+	"net/http"
+	"net/url"
+	"path"
+	"strconv"
+	"sync"
+	"time"
+
+	"hakurei.app/internal/info"
+	"hakurei.app/internal/rosa"
+)
+
+// for lazy initialisation of serveInfo
+var (
+	infoPayload struct {
+		// Current package count.
+		Count int `json:"count"`
+		// Hakurei version, set at link time.
+		HakureiVersion string `json:"hakurei_version"`
+	}
+	infoPayloadOnce sync.Once
+)
+
+// handleInfo writes constant system information.
+func handleInfo(w http.ResponseWriter, _ *http.Request) {
+	infoPayloadOnce.Do(func() {
+		infoPayload.Count = len(rosa.Native().Collect())
+		infoPayload.HakureiVersion = info.Version()
+	})
+	// TODO(mae): cache entire response if no additional fields are planned
+	writeAPIPayload(w, infoPayload)
+}
+
+// newStatusHandler returns a [http.HandlerFunc] that offers status files for
+// viewing or download, if available.
+func (index *packageIndex) newStatusHandler(disposition bool) http.HandlerFunc {
+	return func(w http.ResponseWriter, r *http.Request) {
+		m, ok := index.names[path.Base(r.URL.Path)]
+		if !ok || !m.HasReport {
+			http.NotFound(w, r)
+			return
+		}
+
+		contentType := "text/plain; charset=utf-8"
+		if disposition {
+			contentType = "application/octet-stream"
+
+			// quoting like this is unsound, but okay, because metadata is hardcoded
+			contentDisposition := `attachment; filename="`
+			contentDisposition += m.Name + "-"
+			if m.Version != "" {
+				contentDisposition += m.Version + "-"
+			}
+			contentDisposition += m.ids + `.log"`
+			w.Header().Set("Content-Disposition", contentDisposition)
+		}
+		w.Header().Set("Content-Type", contentType)
+		w.Header().Set("Cache-Control", "no-cache, no-store, must-revalidate")
+		if err := func() (err error) {
+			defer index.handleAccess(&err)()
+			_, err = w.Write(m.status)
+			return
+		}(); err != nil {
+			log.Println(err)
+			http.Error(
+				w, "cannot deliver status, contact maintainers",
+				http.StatusInternalServerError,
+			)
+		}
+	}
+}
+
+// handleGet writes a slice of metadata with specified order.
+func (index *packageIndex) handleGet(w http.ResponseWriter, r *http.Request) {
+	q := r.URL.Query()
+	limit, err := strconv.Atoi(q.Get("limit"))
+	if err != nil || limit > 100 || limit < 1 {
+		http.Error(
+			w, "limit must be an integer between 1 and 100",
+			http.StatusBadRequest,
+		)
+		return
+	}
+	i, err := strconv.Atoi(q.Get("index"))
+	if err != nil || i >= len(index.sorts[0]) || i < 0 {
+		http.Error(
+			w, "index must be an integer between 0 and "+
+				strconv.Itoa(len(index.sorts[0])-1),
+			http.StatusBadRequest,
+		)
+		return
+	}
+	sort, err := strconv.Atoi(q.Get("sort"))
+	if err != nil || sort >= len(index.sorts) || sort < 0 {
+		http.Error(
+			w, "sort must be an integer between 0 and "+
+				strconv.Itoa(sortOrderEnd),
+			http.StatusBadRequest,
+		)
+		return
+	}
+	values := index.sorts[sort][i:min(i+limit, len(index.sorts[sort]))]
+	writeAPIPayload(w, &struct {
+		Values []*metadata `json:"values"`
+	}{values})
+}
+
+func (index *packageIndex) handleSearch(w http.ResponseWriter, r *http.Request) {
+	q := r.URL.Query()
+	limit, err := strconv.Atoi(q.Get("limit"))
+	if err != nil || limit > 100 || limit < 1 {
+		http.Error(
+			w, "limit must be an integer between 1 and 100",
+			http.StatusBadRequest,
+		)
+		return
+	}
+	i, err := strconv.Atoi(q.Get("index"))
+	if err != nil || i >= len(index.sorts[0]) || i < 0 {
+		http.Error(
+			w, "index must be an integer between 0 and "+
+				strconv.Itoa(len(index.sorts[0])-1),
+			http.StatusBadRequest,
+		)
+		return
+	}
+	search, err := url.QueryUnescape(q.Get("search"))
+	if len(search) > 100 || err != nil {
+		http.Error(
+			w, "search must be a string between 0 and 100 characters long",
+			http.StatusBadRequest,
+		)
+		return
+	}
+	desc := q.Get("desc") == "true"
+	n, res, err := index.performSearchQuery(limit, i, search, desc)
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+	}
+	writeAPIPayload(w, &struct {
+		Count  int            `json:"count"`
+		Values []searchResult `json:"values"`
+	}{n, res})
+}
+
+// apiVersion is the name of the current API revision, as part of the pattern.
+const apiVersion = "v1"
+
+// registerAPI registers API handler functions.
+func (index *packageIndex) registerAPI(mux *http.ServeMux) {
+	mux.HandleFunc("GET /api/"+apiVersion+"/info", handleInfo)
+	mux.HandleFunc("GET /api/"+apiVersion+"/get", index.handleGet)
+	mux.HandleFunc("GET /api/"+apiVersion+"/search", index.handleSearch)
+	mux.HandleFunc("GET /api/"+apiVersion+"/status/", index.newStatusHandler(false))
+	mux.HandleFunc("GET /status/", index.newStatusHandler(true))
+}
+
+// Register arranges for mux to service API requests.
+func Register(ctx context.Context, mux *http.ServeMux, report *rosa.Report) error {
+	var index packageIndex
+	index.search = make(searchCache)
+	if err := index.populate(report); err != nil {
+		return err
+	}
+	ticker := time.NewTicker(1 * time.Minute)
+	go func() {
+		for {
+			select {
+			case <-ctx.Done():
+				ticker.Stop()
+				return
+			case <-ticker.C:
+				index.search.clean()
+			}
+		}
+	}()
+	index.registerAPI(mux)
+	return nil
+}
+
+// writeAPIPayload sets headers common to API responses and encodes payload as
+// JSON for the response body.
+func writeAPIPayload(w http.ResponseWriter, payload any) {
+	w.Header().Set("Content-Type", "application/json; charset=utf-8")
+	w.Header().Set("Cache-Control", "no-cache, no-store, must-revalidate")
+	w.Header().Set("Pragma", "no-cache")
+	w.Header().Set("Expires", "0")
+
+	if err := json.NewEncoder(w).Encode(payload); err != nil {
+		log.Println(err)
+		http.Error(
+			w, "cannot encode payload, contact maintainers",
+			http.StatusInternalServerError,
+		)
+	}
+}

cmd/mbf/internal/pkgserver/api_test.go

@@ -0,0 +1,111 @@
+package pkgserver
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"strconv"
+	"testing"
+
+	"hakurei.app/internal/info"
+	"hakurei.app/internal/rosa"
+)
+
+// prefix is prepended to every API path.
+const prefix = "/api/" + apiVersion + "/"
+
+func TestAPIInfo(t *testing.T) {
+	t.Parallel()
+
+	w := httptest.NewRecorder()
+	handleInfo(w, httptest.NewRequestWithContext(
+		t.Context(),
+		http.MethodGet,
+		prefix+"info",
+		nil,
+	))
+
+	resp := w.Result()
+	checkStatus(t, resp, http.StatusOK)
+	checkAPIHeader(t, w.Header())
+
+	checkPayload(t, resp, struct {
+		Count          int    `json:"count"`
+		HakureiVersion string `json:"hakurei_version"`
+	}{len(rosa.Native().Collect()), info.Version()})
+}
+
+func TestAPIGet(t *testing.T) {
+	t.Parallel()
+	const target = prefix + "get"
+
+	index := newIndex(t)
+	newRequest := func(suffix string) *httptest.ResponseRecorder {
+		w := httptest.NewRecorder()
+		index.handleGet(w, httptest.NewRequestWithContext(
+			t.Context(),
+			http.MethodGet,
+			target+suffix,
+			nil,
+		))
+		return w
+	}
+
+	checkValidate := func(t *testing.T, suffix string, vmin, vmax int, wantErr string) {
+		t.Run("invalid", func(t *testing.T) {
+			t.Parallel()
+
+			w := newRequest("?" + suffix + "=invalid")
+			resp := w.Result()
+			checkError(t, resp, wantErr, http.StatusBadRequest)
+		})
+
+		t.Run("min", func(t *testing.T) {
+			t.Parallel()
+
+			w := newRequest("?" + suffix + "=" + strconv.Itoa(vmin-1))
+			resp := w.Result()
+			checkError(t, resp, wantErr, http.StatusBadRequest)
+
+			w = newRequest("?" + suffix + "=" + strconv.Itoa(vmin))
+			resp = w.Result()
+			checkStatus(t, resp, http.StatusOK)
+		})
+
+		t.Run("max", func(t *testing.T) {
+			t.Parallel()
+
+			w := newRequest("?" + suffix + "=" + strconv.Itoa(vmax+1))
+			resp := w.Result()
+			checkError(t, resp, wantErr, http.StatusBadRequest)
+
+			w = newRequest("?" + suffix + "=" + strconv.Itoa(vmax))
+			resp = w.Result()
+			checkStatus(t, resp, http.StatusOK)
+		})
+	}
+
+	t.Run("limit", func(t *testing.T) {
+		t.Parallel()
+		checkValidate(
+			t, "index=0&sort=0&limit", 1, 100,
+			"limit must be an integer between 1 and 100",
+		)
+	})
+
+	count := len(rosa.Native().Collect())
+	t.Run("index", func(t *testing.T) {
+		t.Parallel()
+		checkValidate(
+			t, "limit=1&sort=0&index", 0, count-1,
+			"index must be an integer between 0 and "+strconv.Itoa(count-1),
+		)
+	})
+
+	t.Run("sort", func(t *testing.T) {
+		t.Parallel()
+		checkValidate(
+			t, "index=0&limit=1&sort", 0, int(sortOrderEnd),
+			"sort must be an integer between 0 and "+strconv.Itoa(int(sortOrderEnd)),
+		)
+	})
+}

cmd/mbf/internal/pkgserver/index.go

@@ -0,0 +1,108 @@
+package pkgserver
+
+import (
+	"cmp"
+	"errors"
+	"slices"
+	"strings"
+
+	"hakurei.app/internal/pkg"
+	"hakurei.app/internal/rosa"
+)
+
+const (
+	declarationAscending = iota
+	declarationDescending
+	nameAscending
+	nameDescending
+	sizeAscending
+	sizeDescending
+
+	sortOrderEnd = iota - 1
+)
+
+// packageIndex refers to metadata by name and various sort orders.
+type packageIndex struct {
+	sorts  [sortOrderEnd + 1][]*metadata
+	names  map[string]*metadata
+	search searchCache
+	// Taken from [rosa.Report] if available.
+	handleAccess func(*error) func()
+}
+
+// metadata holds [rosa.Metadata] extended with additional information.
+type metadata struct {
+	handle rosa.ArtifactH
+	*rosa.Metadata
+
+	// Copied from [rosa.Metadata], [rosa.Unversioned] is equivalent to the zero
+	// value. Otherwise, the zero value is invalid.
+	Version string `json:"version,omitempty"`
+	// Output data size, available if present in report.
+	Size int64 `json:"size,omitempty"`
+	// Whether the underlying [pkg.Artifact] is present in the report.
+	HasReport bool `json:"report"`
+
+	// Ident string encoded ahead of time.
+	ids string
+	// Backed by [rosa.Report], access must be prepared by HandleAccess.
+	status []byte
+}
+
+// populate deterministically populates packageIndex, optionally with a report.
+func (index *packageIndex) populate(report *rosa.Report) (err error) {
+	if report != nil {
+		defer report.HandleAccess(&err)()
+		index.handleAccess = report.HandleAccess
+	}
+
+	handles := rosa.Native().Collect()
+	work := make([]*metadata, len(handles))
+	index.names = make(map[string]*metadata)
+	ir := pkg.NewIR()
+	for i, handle := range handles {
+		meta, a := rosa.Native().Std().MustLoad(handle)
+		m := metadata{
+			handle: handle,
+
+			Metadata: meta,
+			Version:  meta.Version,
+		}
+		if m.Version == "" {
+			return errors.New("invalid version from " + m.Name)
+		}
+		if m.Version == rosa.Unversioned {
+			m.Version = ""
+		}
+
+		if report != nil {
+			id := ir.Ident(a)
+			m.ids = pkg.Encode(id.Value())
+			m.status, m.Size = report.ArtifactOf(id)
+			m.HasReport = m.Size >= 0
+		}
+
+		work[i] = &m
+		index.names[m.Name] = &m
+	}
+
+	index.sorts[declarationAscending] = work
+	index.sorts[declarationDescending] = slices.Clone(work)
+	slices.Reverse(index.sorts[declarationDescending][:])
+
+	index.sorts[nameAscending] = slices.Clone(work)
+	slices.SortFunc(index.sorts[nameAscending][:], func(a, b *metadata) int {
+		return strings.Compare(a.Name, b.Name)
+	})
+	index.sorts[nameDescending] = slices.Clone(index.sorts[nameAscending])
+	slices.Reverse(index.sorts[nameDescending][:])
+
+	index.sorts[sizeAscending] = slices.Clone(work)
+	slices.SortFunc(index.sorts[sizeAscending][:], func(a, b *metadata) int {
+		return cmp.Compare(a.Size, b.Size)
+	})
+	index.sorts[sizeDescending] = slices.Clone(index.sorts[sizeAscending])
+	slices.Reverse(index.sorts[sizeDescending][:])
+
+	return
+}

cmd/mbf/internal/pkgserver/index_test.go

@@ -0,0 +1,96 @@
+package pkgserver
+
+import (
+	"bytes"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"reflect"
+	"testing"
+)
+
+// newIndex returns the address of a newly populated packageIndex.
+func newIndex(t *testing.T) *packageIndex {
+	t.Helper()
+
+	var index packageIndex
+	if err := index.populate(nil); err != nil {
+		t.Fatalf("populate: error = %v", err)
+	}
+	return &index
+}
+
+// checkStatus checks response status code.
+func checkStatus(t *testing.T, resp *http.Response, want int) {
+	t.Helper()
+
+	if resp.StatusCode != want {
+		t.Errorf(
+			"StatusCode: %s, want %s",
+			http.StatusText(resp.StatusCode),
+			http.StatusText(want),
+		)
+	}
+}
+
+// checkHeader checks the value of a header entry.
+func checkHeader(t *testing.T, h http.Header, key, want string) {
+	t.Helper()
+
+	if got := h.Get(key); got != want {
+		t.Errorf("%s: %q, want %q", key, got, want)
+	}
+}
+
+// checkAPIHeader checks common entries set for API endpoints.
+func checkAPIHeader(t *testing.T, h http.Header) {
+	t.Helper()
+
+	checkHeader(t, h, "Content-Type", "application/json; charset=utf-8")
+	checkHeader(t, h, "Cache-Control", "no-cache, no-store, must-revalidate")
+	checkHeader(t, h, "Pragma", "no-cache")
+	checkHeader(t, h, "Expires", "0")
+}
+
+// checkPayloadFunc checks the JSON response of an API endpoint by passing it to f.
+func checkPayloadFunc[T any](
+	t *testing.T,
+	resp *http.Response,
+	f func(got *T) bool,
+) {
+	t.Helper()
+
+	var got T
+	r := io.Reader(resp.Body)
+	if testing.Verbose() {
+		var buf bytes.Buffer
+		r = io.TeeReader(r, &buf)
+		defer func() { t.Helper(); t.Log(buf.String()) }()
+	}
+	if err := json.NewDecoder(r).Decode(&got); err != nil {
+		t.Fatalf("Decode: error = %v", err)
+	}
+
+	if !f(&got) {
+		t.Errorf("Body: %#v", got)
+	}
+}
+
+// checkPayload checks the JSON response of an API endpoint.
+func checkPayload[T any](t *testing.T, resp *http.Response, want T) {
+	t.Helper()
+
+	checkPayloadFunc(t, resp, func(got *T) bool {
+		return reflect.DeepEqual(got, &want)
+	})
+}
+
+func checkError(t *testing.T, resp *http.Response, error string, code int) {
+	t.Helper()
+
+	checkStatus(t, resp, code)
+	if got, _ := io.ReadAll(resp.Body); string(got) != fmt.Sprintln(error) {
+		t.Errorf("Body: %q, want %q", string(got), error)
+	}
+}

cmd/mbf/internal/pkgserver/search.go

@@ -0,0 +1,81 @@
+package pkgserver
+
+import (
+	"cmp"
+	"maps"
+	"regexp"
+	"slices"
+	"time"
+)
+
+type searchCache map[string]searchCacheEntry
+type searchResult struct {
+	NameIndices [][]int `json:"name_matches"`
+	DescIndices [][]int `json:"desc_matches,omitempty"`
+	Score       float64 `json:"score"`
+	*metadata
+}
+type searchCacheEntry struct {
+	query   string
+	results []searchResult
+	expiry  time.Time
+}
+
+func (index *packageIndex) performSearchQuery(limit int, i int, search string, desc bool) (int, []searchResult, error) {
+	query := search
+	if desc {
+		query += ";withDesc"
+	}
+	entry, ok := index.search[query]
+	if ok && len(entry.results) > 0 {
+		return len(entry.results), entry.results[min(i, len(entry.results)-1):min(i+limit, len(entry.results))], nil
+	}
+
+	regex, err := regexp.Compile(search)
+	if err != nil {
+		return 0, make([]searchResult, 0), err
+	}
+	res := make([]searchResult, 0)
+	for p := range maps.Values(index.names) {
+		nameIndices := regex.FindAllIndex([]byte(p.Name), -1)
+		var descIndices [][]int = nil
+		if desc {
+			descIndices = regex.FindAllIndex([]byte(p.Description), -1)
+		}
+		if nameIndices == nil && descIndices == nil {
+			continue
+		}
+		score := float64(indexsum(nameIndices)) / (float64(len(nameIndices)) + 1)
+		if desc {
+			score += float64(indexsum(descIndices)) / (float64(len(descIndices)) + 1) / 10.0
+		}
+		res = append(res, searchResult{
+			NameIndices: nameIndices,
+			DescIndices: descIndices,
+			Score:       score,
+			metadata:    p,
+		})
+	}
+	slices.SortFunc(res[:], func(a, b searchResult) int { return -cmp.Compare(a.Score, b.Score) })
+	expiry := time.Now().Add(1 * time.Minute)
+	entry = searchCacheEntry{
+		query:   search,
+		results: res,
+		expiry:  expiry,
+	}
+	index.search[query] = entry
+
+	return len(res), res[i:min(i+limit, len(entry.results))], nil
+}
+func (s *searchCache) clean() {
+	maps.DeleteFunc(*s, func(_ string, v searchCacheEntry) bool {
+		return v.expiry.Before(time.Now())
+	})
+}
+func indexsum(in [][]int) int {
+	sum := 0
+	for i := range in {
+		sum += in[i][1] - in[i][0]
+	}
+	return sum
+}

cmd/mbf/internal/pkgserver/ui/index.html

@@ -0,0 +1,58 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <link rel="stylesheet" href="style.css">
+    <link rel="icon" href="https://hakurei.app/favicon.ico"/>
+    <title>Rosa OS Packages</title>
+    <script src="index.js"></script>
+</head>
+<body>
+<h1>Rosa OS Packages</h1>
+<div class="top-controls" id="top-controls-regular">
+    <p>Showing entries <span id="entry-counter"></span>.</p>
+    <span id="search-bar">
+        <label for="search">Search: </label>
+        <input type="text" name="search" id="search"/>
+        <button onclick="doSearch()">Find</button>
+        <label for="include-desc">Include descriptions: </label>
+        <input type="checkbox" name="include-desc" id="include-desc" checked/>
+    </span>
+    <div><label for="count">Entries per page: </label><select name="count" id="count">
+        <option value="10">10</option>
+        <option value="20">20</option>
+        <option value="30">30</option>
+        <option value="50">50</option>
+    </select></div>
+    <div><label for="sort">Sort by: </label><select name="sort" id="sort">
+        <option value="0">Definition (ascending)</option>
+        <option value="1">Definition (descending)</option>
+        <option value="2">Name (ascending)</option>
+        <option value="3">Name (descending)</option>
+        <option value="4">Size (ascending)</option>
+        <option value="5">Size (descending)</option>
+    </select></div>
+</div>
+<div class="top-controls" id="search-top-controls" hidden>
+    <p>Showing search results <span id="search-entry-counter"></span> for query "<span id="search-query"></span>".</p>
+    <button onclick="exitSearch()">Back</button>
+    <div><label for="search-count">Entries per page: </label><select name="search-count" id="search-count">
+        <option value="10">10</option>
+        <option value="20">20</option>
+        <option value="30">30</option>
+        <option value="50">50</option>
+    </select></div>
+    <p>Sorted by best match</p>
+</div>
+<div class="page-controls"><a href="javascript:prevPage()">&laquo; Previous</a> <input type="text" class="page-number" value="1"/> <a href="javascript:nextPage()">Next &raquo;</a></div>
+<table id="pkg-list">
+    <tr><td>Loading...</td></tr>
+</table>
+<div class="page-controls"><a href="javascript:prevPage()">&laquo; Previous</a> <input type="text" class="page-number" value="1"/> <a href="javascript:nextPage()">Next &raquo;</a></div>
+<footer>
+    <p>&copy;<a href="https://hakurei.app/">Hakurei</a> (<span id="hakurei-version">unknown</span>). Licensed under the MIT license.</p>
+</footer>
+<script>main();</script>
+</body>
+</html>

cmd/mbf/internal/pkgserver/ui/index.ts

@@ -0,0 +1,331 @@
+interface PackageIndexEntry {
+    name: string
+    size?: number
+    description?: string
+    website?: string
+    version?: string
+    report?: boolean
+}
+
+function entryToHTML(entry: PackageIndexEntry | SearchResult): HTMLTableRowElement {
+    let v = entry.version != null ? `<span>${escapeHtml(entry.version)}</span>` : ""
+    let s = entry.size != null && entry.size > 0 ? `<p>Size: ${toByteSizeString(entry.size)} (${entry.size})</p>` : ""
+    let n: string
+    let d: string
+    if ('name_matches' in entry) {
+        n = `<h2>${nameMatches(entry as SearchResult)} ${v}</h2>`
+    } else {
+        n = `<h2>${escapeHtml(entry.name)} ${v}</h2>`
+    }
+    if ('desc_matches' in entry && STATE.getIncludeDescriptions()) {
+        d = descMatches(entry as SearchResult)
+    } else {
+        d = (entry as PackageIndexEntry).description != null ? `<p>${escapeHtml((entry as PackageIndexEntry).description)}</p>` : ""
+    }
+    let w = entry.website != null ? `<a href="${encodeURI(entry.website)}">Website</a>` : ""
+    let r = entry.report ? `Log (<a href=\"${encodeURI('/api/v1/status/' + entry.name)}\">View</a> | <a href=\"${encodeURI('/status/' + entry.name)}\">Download</a>)` : ""
+    let row = <HTMLTableRowElement>(document.createElement('tr'))
+    row.innerHTML = `<td>
+            ${n}
+            ${d}
+            ${s}
+            ${w}
+            ${r}
+        </td>`
+    return row
+}
+
+function nameMatches(sr: SearchResult): string {
+    return markMatches(sr.name, sr.name_matches)
+}
+
+function descMatches(sr: SearchResult): string {
+    return markMatches(sr.description!, sr.desc_matches)
+}
+
+function markMatches(str: string, indices: [number, number][]): string {
+    if (indices == null) {
+        return str
+    }
+    let out: string = ""
+    let j = 0
+    for (let i = 0; i < str.length; i++) {
+        if (j < indices.length) {
+            if (i === indices[j][0]) {
+                out += `<mark>${escapeHtmlChar(str[i])}`
+                continue
+            }
+            if (i === indices[j][1]) {
+                out += `</mark>${escapeHtmlChar(str[i])}`
+                j++
+                continue
+            }
+        }
+        out += escapeHtmlChar(str[i])
+    }
+    if (indices[j] !== undefined) {
+        out += "</mark>"
+    }
+    return out
+}
+
+function toByteSizeString(bytes: number): string {
+    if (bytes == null) return `unspecified`
+    if (bytes < 1024) return `${bytes}B`
+    if (bytes < Math.pow(1024, 2)) return `${(bytes / 1024).toFixed(2)}kiB`
+    if (bytes < Math.pow(1024, 3)) return `${(bytes / Math.pow(1024, 2)).toFixed(2)}MiB`
+    if (bytes < Math.pow(1024, 4)) return `${(bytes / Math.pow(1024, 3)).toFixed(2)}GiB`
+    if (bytes < Math.pow(1024, 5)) return `${(bytes / Math.pow(1024, 4)).toFixed(2)}TiB`
+    return "not only is it big, it's large"
+}
+
+const API_VERSION = 1
+const ENDPOINT = `/api/v${API_VERSION}`
+
+interface InfoPayload {
+    count?: number
+    hakurei_version?: string
+}
+
+async function infoRequest(): Promise<InfoPayload> {
+    const res = await fetch(`${ENDPOINT}/info`)
+    const payload = await res.json()
+    return payload as InfoPayload
+}
+
+interface GetPayload {
+    values?: PackageIndexEntry[]
+}
+
+enum SortOrders {
+    DeclarationAscending,
+    DeclarationDescending,
+    NameAscending,
+    NameDescending
+}
+
+async function getRequest(limit: number, index: number, sort: SortOrders): Promise<GetPayload> {
+    const res = await fetch(`${ENDPOINT}/get?limit=${limit}&index=${index}&sort=${sort.valueOf()}`)
+    const payload = await res.json()
+    return payload as GetPayload
+}
+
+interface SearchResult extends PackageIndexEntry {
+    name_matches: [number, number][]
+    desc_matches: [number, number][]
+    score: number
+}
+
+interface SearchPayload {
+    count?: number
+    values?: SearchResult[]
+}
+
+async function searchRequest(limit: number, index: number, search: string, desc: boolean): Promise<SearchPayload> {
+    const res = await fetch(`${ENDPOINT}/search?limit=${limit}&index=${index}&search=${encodeURIComponent(search)}&desc=${desc}`)
+    if (!res.ok) {
+        exitSearch()
+        alert("invalid search query!")
+        return Promise.reject(res.statusText)
+    }
+    const payload = await res.json()
+    return payload as SearchPayload
+}
+
+class State {
+    entriesPerPage: number = 10
+    entryIndex: number = 0
+    maxTotal: number = 0
+    maxEntries: number = 0
+    sort: SortOrders = SortOrders.DeclarationAscending
+    search: boolean = false
+
+    getEntriesPerPage(): number {
+        return this.entriesPerPage
+    }
+
+    setEntriesPerPage(entriesPerPage: number) {
+        this.entriesPerPage = entriesPerPage
+        this.setEntryIndex(Math.floor(this.getEntryIndex() / entriesPerPage) * entriesPerPage)
+    }
+
+    getEntryIndex(): number {
+        return this.entryIndex
+    }
+
+    setEntryIndex(entryIndex: number) {
+        this.entryIndex = entryIndex
+        this.updatePage()
+        this.updateRange()
+        this.updateListings()
+    }
+
+    getMaxTotal(): number {
+        return this.maxTotal
+    }
+
+    setMaxTotal(max: number) {
+        this.maxTotal = max
+    }
+
+    getSortOrder(): SortOrders {
+        return this.sort
+    }
+
+    setSortOrder(sortOrder: SortOrders) {
+        this.sort = sortOrder
+        this.setEntryIndex(0)
+    }
+
+    updatePage() {
+        let page = Math.ceil(((this.getEntryIndex() + this.getEntriesPerPage()) - 1) / this.getEntriesPerPage())
+        for (let e of document.getElementsByClassName("page-number")) {
+            (e as HTMLInputElement).value = String(page)
+        }
+    }
+
+    updateRange() {
+        let max = Math.min(this.getEntryIndex() + this.getEntriesPerPage(), this.getMaxTotal())
+        document.getElementById("entry-counter")!.textContent = `${this.getEntryIndex() + 1}-${max} of ${this.getMaxTotal()}`
+        if (this.search) {
+            document.getElementById("search-entry-counter")!.textContent = `${this.getEntryIndex() + 1}-${max} of ${this.maxTotal}/${this.maxEntries}`
+            document.getElementById("search-query")!.innerHTML = `<code>${escapeHtml(this.getSearchQuery())}</code>`
+        }
+    }
+
+    getSearchQuery(): string {
+        let queryString = document.getElementById("search")!;
+        return (queryString as HTMLInputElement).value
+    }
+
+    getIncludeDescriptions(): boolean {
+        let includeDesc = document.getElementById("include-desc")!;
+        return (includeDesc as HTMLInputElement).checked
+    }
+
+    updateListings() {
+        if (this.search) {
+            searchRequest(this.getEntriesPerPage(), this.getEntryIndex(), this.getSearchQuery(), this.getIncludeDescriptions())
+                .then(res => {
+                    let table = document.getElementById("pkg-list")!
+                    table.innerHTML = ''
+                    for (let row of res.values!) {
+                        table.appendChild(entryToHTML(row))
+                    }
+                    STATE.maxTotal = res.count!
+                    STATE.updateRange()
+                    if(res.count! < 1) {
+                        exitSearch()
+                        alert("no results found!")
+                    }
+                })
+        } else {
+            getRequest(this.getEntriesPerPage(), this.getEntryIndex(), this.getSortOrder())
+                .then(res => {
+                    let table = document.getElementById("pkg-list")!
+                    table.innerHTML = ''
+                    for (let row of res.values!) {
+                        table.appendChild(entryToHTML(row))
+                    }
+                })
+        }
+    }
+}
+
+let STATE: State
+
+
+function lastPageIndex(): number {
+    return Math.floor(STATE.getMaxTotal() / STATE.getEntriesPerPage()) * STATE.getEntriesPerPage()
+}
+
+function setPage(page: number) {
+    STATE.setEntryIndex(Math.max(0, Math.min(STATE.getEntriesPerPage() * (page - 1), lastPageIndex())))
+}
+
+
+function escapeHtml(str?: string): string {
+    let out: string = ''
+    if (str == undefined) return ""
+    for (let i = 0; i < str.length; i++) {
+        out += escapeHtmlChar(str[i])
+    }
+    return out
+}
+
+function escapeHtmlChar(char: string): string {
+    if (char.length != 1) return char
+    switch (char[0]) {
+        case '&':
+            return "&amp;"
+        case '<':
+            return "&lt;"
+        case '>':
+            return "&gt;"
+        case '"':
+            return "&quot;"
+        case "'":
+            return "&apos;"
+        default:
+            return char
+    }
+}
+
+function firstPage() {
+    STATE.setEntryIndex(0)
+}
+
+function prevPage() {
+    let index = STATE.getEntryIndex()
+    STATE.setEntryIndex(Math.max(0, index - STATE.getEntriesPerPage()))
+}
+
+function lastPage() {
+    STATE.setEntryIndex(lastPageIndex())
+}
+
+function nextPage() {
+    let index = STATE.getEntryIndex()
+    STATE.setEntryIndex(Math.min(lastPageIndex(), index + STATE.getEntriesPerPage()))
+}
+
+function doSearch() {
+    document.getElementById("top-controls-regular")!.toggleAttribute("hidden");
+    document.getElementById("search-top-controls")!.toggleAttribute("hidden");
+    STATE.search = true;
+    STATE.setEntryIndex(0);
+}
+
+function exitSearch() {
+    document.getElementById("top-controls-regular")!.toggleAttribute("hidden");
+    document.getElementById("search-top-controls")!.toggleAttribute("hidden");
+    STATE.search = false;
+    STATE.setMaxTotal(STATE.maxEntries)
+    STATE.setEntryIndex(0)
+}
+
+function main() {
+    STATE = new State()
+    infoRequest()
+        .then(res => {
+            STATE.maxEntries = res.count!
+            STATE.setMaxTotal(STATE.maxEntries)
+            document.getElementById("hakurei-version")!.textContent = res.hakurei_version!
+            STATE.updateRange()
+            STATE.updateListings()
+        })
+    for (let e of document.getElementsByClassName("page-number")) {
+        e.addEventListener("change", (_) => {
+            setPage(parseInt((e as HTMLInputElement).value))
+        })
+    }
+    document.getElementById("count")?.addEventListener("change", (event) => {
+        STATE.setEntriesPerPage(parseInt((event.target as HTMLSelectElement).value))
+    })
+    document.getElementById("sort")?.addEventListener("change", (event) => {
+        STATE.setSortOrder(parseInt((event.target as HTMLSelectElement).value))
+    })
+    document.getElementById("search")?.addEventListener("keyup", (event) => {
+        if (event.key === 'Enter') doSearch()
+    })
+}

cmd/mbf/internal/pkgserver/ui/style.css

@@ -0,0 +1,21 @@
+.page-number {
+  width: 2em;
+  text-align: center;
+}
+.page-number {
+  width: 2em;
+  text-align: center;
+}
+
+@media (prefers-color-scheme: dark) {
+  html {
+    background-color: #2c2c2c;
+    color: ghostwhite;
+  }
+}
+@media (prefers-color-scheme: light) {
+  html {
+    background-color: #d3d3d3;
+    color: black;
+  }
+}

cmd/mbf/internal/pkgserver/ui/tsconfig.json

@@ -0,0 +1,8 @@
+{
+  "compilerOptions": {
+    "target": "ES2024",
+    "strict": true,
+    "alwaysStrict": true,
+    "outDir": "static"
+  }
+}

cmd/mbf/internal/pkgserver/ui/ui.go

@@ -0,0 +1,9 @@
+// Package ui holds the static web UI.
+package ui
+
+import "net/http"
+
+// Register arranges for mux to serve the embedded frontend.
+func Register(mux *http.ServeMux) {
+	mux.Handle("GET /", http.FileServer(http.FS(static)))
+}

cmd/mbf/internal/pkgserver/ui/ui_full.go

@@ -0,0 +1,21 @@
+//go:build frontend
+
+package ui
+
+import (
+	"embed"
+	"io/fs"
+)
+
+//go:generate tsc
+//go:generate cp index.html style.css static
+//go:embed static
+var _static embed.FS
+
+var static = func() fs.FS {
+	if f, err := fs.Sub(_static, "static"); err != nil {
+		panic(err)
+	} else {
+		return f
+	}
+}()

cmd/mbf/internal/pkgserver/ui/ui_stub.go

@@ -0,0 +1,7 @@
+//go:build !frontend
+
+package ui
+
+import "testing/fstest"
+
+var static fstest.MapFS

cmd/mbf/main_test.go

@@ -0,0 +1,47 @@
+package main
+
+import (
+	"net"
+	"os"
+	"testing"
+
+	"hakurei.app/internal/rosa"
+)
+
+func TestMain(m *testing.M) {
+	rosa.Native().DropCaches("", rosa.OptLLVMNoLTO)
+	os.Exit(m.Run())
+}
+
+func TestCureAll(t *testing.T) {
+	t.Parallel()
+	const env = "ROSA_TEST_DAEMON"
+
+	if !testing.Verbose() {
+		t.Skip("verbose flag not set")
+	}
+
+	pathname, ok := os.LookupEnv(env)
+	if !ok {
+		t.Skip(env + " not set")
+	}
+
+	addr := net.UnixAddr{Net: "unix", Name: pathname}
+	t.Cleanup(func() {
+		if t.Failed() {
+			if err := abortRemote(t.Context(), &addr, false); err != nil {
+				t.Fatal(err)
+			}
+		}
+	})
+
+	for _, handle := range rosa.Native().Collect() {
+		_, a := rosa.Native().Std().MustLoad(handle)
+		t.Run(handle.String(), func(t *testing.T) {
+			_, err := cureRemote(t.Context(), &addr, a, 0)
+			if err != nil {
+				t.Error(err)
+			}
+		})
+	}
+}

cmd/sharefs/fuse-operations.c

@@ -0,0 +1,282 @@
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE /* O_DIRECT */
+#endif
+
+#include <dirent.h>
+#include <errno.h>
+#include <unistd.h>
+
+/* TODO(ophestra): remove after 05ce67fea99ca09cd4b6625cff7aec9cc222dd5a reaches a release */
+#include <sys/syscall.h>
+
+#include "fuse-operations.h"
+
+/* MUST_TRANSLATE_PATHNAME translates a userspace pathname to a relative pathname;
+ * the resulting address points to a constant string or part of pathname, it is never heap allocated. */
+#define MUST_TRANSLATE_PATHNAME(pathname) \
+    do {                                  \
+        if (pathname == NULL)             \
+            return -EINVAL;               \
+        while (*pathname == '/')          \
+            pathname++;                   \
+        if (*pathname == '\0')            \
+            pathname = ".";               \
+    } while (0)
+
+/* GET_CONTEXT_PRIV obtains fuse context and private data for the calling thread. */
+#define GET_CONTEXT_PRIV(ctx, priv) \
+    do {                            \
+        ctx = fuse_get_context();   \
+        priv = ctx->private_data;   \
+    } while (0)
+
+/* impl_getattr modifies a struct stat from the kernel to present to userspace;
+ * impl_getattr returns a negative errno style error code. */
+static int impl_getattr(struct fuse_context *ctx, struct stat *statbuf) {
+    /* allowlist of permitted types */
+    if (!S_ISDIR(statbuf->st_mode) && !S_ISREG(statbuf->st_mode) && !S_ISLNK(statbuf->st_mode)) {
+        return -ENOTRECOVERABLE; /* returning an errno causes all operations on the file to return EIO */
+    }
+
+#define OVERRIDE_PERM(v) (statbuf->st_mode & ~0777) | (v & 0777)
+    if (S_ISDIR(statbuf->st_mode))
+        statbuf->st_mode = OVERRIDE_PERM(SHAREFS_PERM_DIR);
+    else if (S_ISREG(statbuf->st_mode))
+        statbuf->st_mode = OVERRIDE_PERM(SHAREFS_PERM_REG);
+    else
+        statbuf->st_mode = 0; /* should always be symlink in this case */
+
+    statbuf->st_uid = ctx->uid;
+    statbuf->st_gid = SHAREFS_MEDIA_RW_ID;
+    statbuf->st_ctim = statbuf->st_mtim;
+    statbuf->st_nlink = 1;
+    return 0;
+}
+
+/* fuse_operations implementation */
+
+int sharefs_getattr(const char *pathname, struct stat *statbuf, struct fuse_file_info *fi) {
+    struct fuse_context *ctx;
+    struct sharefs_private *priv;
+    GET_CONTEXT_PRIV(ctx, priv);
+    MUST_TRANSLATE_PATHNAME(pathname);
+
+    (void)fi;
+
+    if (fstatat(priv->dirfd, pathname, statbuf, AT_SYMLINK_NOFOLLOW) == -1)
+        return -errno;
+    return impl_getattr(ctx, statbuf);
+}
+
+int sharefs_readdir(const char *pathname, void *buf, fuse_fill_dir_t filler, off_t offset, struct fuse_file_info *fi, enum fuse_readdir_flags flags) {
+    int fd;
+    DIR *dp;
+    struct stat st;
+    int ret = 0;
+    struct dirent *de;
+
+    struct fuse_context *ctx;
+    struct sharefs_private *priv;
+    GET_CONTEXT_PRIV(ctx, priv);
+    MUST_TRANSLATE_PATHNAME(pathname);
+
+    (void)offset;
+    (void)fi;
+
+    if ((fd = openat(priv->dirfd, pathname, O_RDONLY | O_DIRECTORY | O_CLOEXEC)) == -1)
+        return -errno;
+    if ((dp = fdopendir(fd)) == NULL) {
+        close(fd);
+        return -errno;
+    }
+
+    errno = 0; /* for the next readdir call */
+    while ((de = readdir(dp)) != NULL) {
+        if (flags & FUSE_READDIR_PLUS) {
+            if (fstatat(dirfd(dp), de->d_name, &st, AT_SYMLINK_NOFOLLOW) == -1) {
+                ret = -errno;
+                break;
+            }
+
+            if ((ret = impl_getattr(ctx, &st)) < 0)
+                break;
+
+            errno = 0;
+            ret = filler(buf, de->d_name, &st, 0, FUSE_FILL_DIR_PLUS);
+        } else
+            ret = filler(buf, de->d_name, NULL, 0, 0);
+
+        if (ret != 0) {
+            ret = errno != 0 ? -errno : -EIO; /* filler */
+            break;
+        }
+
+        errno = 0; /* for the next readdir call */
+    }
+    if (ret == 0 && errno != 0)
+        ret = -errno; /* readdir */
+
+    closedir(dp);
+    return ret;
+}
+
+int sharefs_mkdir(const char *pathname, mode_t mode) {
+    struct fuse_context *ctx;
+    struct sharefs_private *priv;
+    GET_CONTEXT_PRIV(ctx, priv);
+    MUST_TRANSLATE_PATHNAME(pathname);
+
+    (void)mode;
+
+    if (mkdirat(priv->dirfd, pathname, SHAREFS_PERM_DIR) == -1)
+        return -errno;
+    return 0;
+}
+
+int sharefs_unlink(const char *pathname) {
+    struct fuse_context *ctx;
+    struct sharefs_private *priv;
+    GET_CONTEXT_PRIV(ctx, priv);
+    MUST_TRANSLATE_PATHNAME(pathname);
+
+    if (unlinkat(priv->dirfd, pathname, 0) == -1)
+        return -errno;
+    return 0;
+}
+
+int sharefs_rmdir(const char *pathname) {
+    struct fuse_context *ctx;
+    struct sharefs_private *priv;
+    GET_CONTEXT_PRIV(ctx, priv);
+    MUST_TRANSLATE_PATHNAME(pathname);
+
+    if (unlinkat(priv->dirfd, pathname, AT_REMOVEDIR) == -1)
+        return -errno;
+    return 0;
+}
+
+int sharefs_rename(const char *oldpath, const char *newpath, unsigned int flags) {
+    struct fuse_context *ctx;
+    struct sharefs_private *priv;
+    GET_CONTEXT_PRIV(ctx, priv);
+    MUST_TRANSLATE_PATHNAME(oldpath);
+    MUST_TRANSLATE_PATHNAME(newpath);
+
+    /* TODO(ophestra): replace with wrapper after 05ce67fea99ca09cd4b6625cff7aec9cc222dd5a reaches a release */
+    if (syscall(__NR_renameat2, priv->dirfd, oldpath, priv->dirfd, newpath, flags) == -1)
+        return -errno;
+    return 0;
+}
+
+int sharefs_truncate(const char *pathname, off_t length, struct fuse_file_info *fi) {
+    int fd;
+    int ret;
+
+    struct fuse_context *ctx;
+    struct sharefs_private *priv;
+    GET_CONTEXT_PRIV(ctx, priv);
+    MUST_TRANSLATE_PATHNAME(pathname);
+
+    (void)fi;
+
+    if ((fd = openat(priv->dirfd, pathname, O_WRONLY | O_CLOEXEC)) == -1)
+        return -errno;
+    if ((ret = ftruncate(fd, length)) == -1)
+        ret = -errno;
+    close(fd);
+    return ret;
+}
+
+int sharefs_utimens(const char *pathname, const struct timespec times[2], struct fuse_file_info *fi) {
+    struct fuse_context *ctx;
+    struct sharefs_private *priv;
+    GET_CONTEXT_PRIV(ctx, priv);
+    MUST_TRANSLATE_PATHNAME(pathname);
+
+    (void)fi;
+
+    if (utimensat(priv->dirfd, pathname, times, AT_SYMLINK_NOFOLLOW) == -1)
+        return -errno;
+    return 0;
+}
+
+int sharefs_create(const char *pathname, mode_t mode, struct fuse_file_info *fi) {
+    int fd;
+
+    struct fuse_context *ctx;
+    struct sharefs_private *priv;
+    GET_CONTEXT_PRIV(ctx, priv);
+    MUST_TRANSLATE_PATHNAME(pathname);
+
+    (void)mode;
+
+    if ((fd = openat(priv->dirfd, pathname, fi->flags & ~SHAREFS_FORBIDDEN_FLAGS, SHAREFS_PERM_REG)) == -1)
+        return -errno;
+    fi->fh = fd;
+    return 0;
+}
+
+int sharefs_open(const char *pathname, struct fuse_file_info *fi) {
+    int fd;
+
+    struct fuse_context *ctx;
+    struct sharefs_private *priv;
+    GET_CONTEXT_PRIV(ctx, priv);
+    MUST_TRANSLATE_PATHNAME(pathname);
+
+    if ((fd = openat(priv->dirfd, pathname, fi->flags & ~SHAREFS_FORBIDDEN_FLAGS)) == -1)
+        return -errno;
+    fi->fh = fd;
+    return 0;
+}
+
+int sharefs_read(const char *pathname, char *buf, size_t count, off_t offset, struct fuse_file_info *fi) {
+    int ret;
+
+    (void)pathname;
+
+    if ((ret = pread(fi->fh, buf, count, offset)) == -1)
+        return -errno;
+    return ret;
+}
+
+int sharefs_write(const char *pathname, const char *buf, size_t count, off_t offset, struct fuse_file_info *fi) {
+    int ret;
+
+    (void)pathname;
+
+    if ((ret = pwrite(fi->fh, buf, count, offset)) == -1)
+        return -errno;
+    return ret;
+}
+
+int sharefs_statfs(const char *pathname, struct statvfs *statbuf) {
+    int fd;
+    int ret;
+
+    struct fuse_context *ctx;
+    struct sharefs_private *priv;
+    GET_CONTEXT_PRIV(ctx, priv);
+    MUST_TRANSLATE_PATHNAME(pathname);
+
+    if ((fd = openat(priv->dirfd, pathname, O_RDONLY | O_CLOEXEC)) == -1)
+        return -errno;
+    if ((ret = fstatvfs(fd, statbuf)) == -1)
+        ret = -errno;
+    close(fd);
+    return ret;
+}
+
+int sharefs_release(const char *pathname, struct fuse_file_info *fi) {
+    (void)pathname;
+
+    return close(fi->fh);
+}
+
+int sharefs_fsync(const char *pathname, int datasync, struct fuse_file_info *fi) {
+    (void)pathname;
+
+    if (datasync ? fdatasync(fi->fh) : fsync(fi->fh) == -1)
+        return -errno;
+    return 0;
+}

cmd/sharefs/fuse-operations.h

@@ -0,0 +1,34 @@
+#define FUSE_USE_VERSION FUSE_MAKE_VERSION(3, 12)
+#include <fuse.h>
+#include <fuse_lowlevel.h> /* for fuse_cmdline_help */
+
+#if (FUSE_VERSION < FUSE_MAKE_VERSION(3, 12))
+#error This package requires libfuse >= v3.12
+#endif
+
+#define SHAREFS_MEDIA_RW_ID (1 << 10) - 1 /* owning gid presented to userspace */
+#define SHAREFS_PERM_DIR 0770             /* permission bits for directories presented to userspace */
+#define SHAREFS_PERM_REG 0660             /* permission bits for regular files presented to userspace */
+#define SHAREFS_FORBIDDEN_FLAGS O_DIRECT  /* these open flags are cleared unconditionally */
+
+/* sharefs_private is populated by sharefs_init and contains process-wide context */
+struct sharefs_private {
+    int dirfd;       /* source dirfd opened during sharefs_init */
+    uintptr_t setup; /* cgo handle of opaque setup state */
+};
+
+int sharefs_getattr(const char *pathname, struct stat *statbuf, struct fuse_file_info *fi);
+int sharefs_readdir(const char *pathname, void *buf, fuse_fill_dir_t filler, off_t offset, struct fuse_file_info *fi, enum fuse_readdir_flags flags);
+int sharefs_mkdir(const char *pathname, mode_t mode);
+int sharefs_unlink(const char *pathname);
+int sharefs_rmdir(const char *pathname);
+int sharefs_rename(const char *oldpath, const char *newpath, unsigned int flags);
+int sharefs_truncate(const char *pathname, off_t length, struct fuse_file_info *fi);
+int sharefs_utimens(const char *pathname, const struct timespec times[2], struct fuse_file_info *fi);
+int sharefs_create(const char *pathname, mode_t mode, struct fuse_file_info *fi);
+int sharefs_open(const char *pathname, struct fuse_file_info *fi);
+int sharefs_read(const char *pathname, char *buf, size_t count, off_t offset, struct fuse_file_info *fi);
+int sharefs_write(const char *pathname, const char *buf, size_t count, off_t offset, struct fuse_file_info *fi);
+int sharefs_statfs(const char *pathname, struct statvfs *statbuf);
+int sharefs_release(const char *pathname, struct fuse_file_info *fi);
+int sharefs_fsync(const char *pathname, int datasync, struct fuse_file_info *fi);

cmd/sharefs/fuse.go

@@ -0,0 +1,570 @@
+package main
+
+/*
+#cgo pkg-config: --static fuse3
+
+#include "fuse-operations.h"
+#include <stdlib.h>
+#include <string.h>
+
+extern void *sharefs_init(struct fuse_conn_info *conn, struct fuse_config *cfg);
+extern void sharefs_destroy(void *private_data);
+
+typedef void (*closure)();
+static inline struct fuse_opt _FUSE_OPT_END() { return (struct fuse_opt)FUSE_OPT_END; };
+*/
+import "C"
+import (
+	"context"
+	"encoding/gob"
+	"errors"
+	"fmt"
+	"log"
+	"os"
+	"os/exec"
+	"os/signal"
+	"path/filepath"
+	"runtime"
+	"runtime/cgo"
+	"strconv"
+	"syscall"
+	"unsafe"
+
+	"hakurei.app/check"
+	"hakurei.app/container"
+	"hakurei.app/container/std"
+	"hakurei.app/fhs"
+	"hakurei.app/hst"
+	"hakurei.app/internal/helper/proc"
+	"hakurei.app/internal/info"
+	"hakurei.app/message"
+)
+
+type (
+	// closure represents a C function pointer.
+	closure = C.closure
+
+	// fuseArgs represents the fuse_args structure.
+	fuseArgs = C.struct_fuse_args
+
+	// setupState holds state used for setup. Its cgo handle is included in
+	// sharefs_private and considered opaque to non-setup callbacks.
+	setupState struct {
+		// Whether sharefs_init failed.
+		initFailed bool
+
+		// Whether to create source directory as root.
+		mkdir bool
+
+		// Open file descriptor to fuse.
+		Fuse int
+
+		// Pathname to open for dirfd.
+		Source *check.Absolute
+		// New uid and gid to set by sharefs_init when starting as root.
+		Setuid, Setgid int
+	}
+)
+
+func init() { gob.Register(new(setupState)) }
+
+// destroySetup invalidates the setup [cgo.Handle] in a sharefs_private structure.
+func destroySetup(private_data unsafe.Pointer) (ok bool) {
+	if private_data == nil {
+		return false
+	}
+	priv := (*C.struct_sharefs_private)(private_data)
+
+	if h := cgo.Handle(priv.setup); h != 0 {
+		priv.setup = 0
+		h.Delete()
+		ok = true
+	}
+	return
+}
+
+//export sharefs_init
+func sharefs_init(
+	_ *C.struct_fuse_conn_info,
+	cfg *C.struct_fuse_config,
+) unsafe.Pointer {
+	ctx := C.fuse_get_context()
+	priv := (*C.struct_sharefs_private)(ctx.private_data)
+	setup := cgo.Handle(priv.setup).Value().(*setupState)
+
+	if os.Geteuid() == 0 {
+		log.Println("filesystem daemon must not run as root")
+		goto fail
+	}
+
+	cfg.use_ino = C.true
+	cfg.direct_io = C.false
+	// getattr is context-dependent
+	cfg.attr_timeout = 0
+	cfg.entry_timeout = 0
+	cfg.negative_timeout = 0
+
+	// all future filesystem operations happen through this dirfd
+	if fd, err := syscall.Open(
+		setup.Source.String(),
+		syscall.O_DIRECTORY|syscall.O_RDONLY|syscall.O_CLOEXEC,
+		0,
+	); err != nil {
+		log.Printf("cannot open %q: %v", setup.Source, err)
+		goto fail
+	} else if err = syscall.Fchdir(fd); err != nil {
+		_ = syscall.Close(fd)
+		log.Printf("cannot enter %q: %s", setup.Source, err)
+		goto fail
+	} else {
+		priv.dirfd = C.int(fd)
+	}
+
+	return ctx.private_data
+
+fail:
+	setup.initFailed = true
+	C.fuse_exit(ctx.fuse)
+	return nil
+}
+
+//export sharefs_destroy
+func sharefs_destroy(private_data unsafe.Pointer) {
+	if private_data != nil {
+		destroySetup(private_data)
+		priv := (*C.struct_sharefs_private)(private_data)
+
+		if err := syscall.Close(int(priv.dirfd)); err != nil {
+			log.Printf("cannot close source directory: %v", err)
+		}
+	}
+}
+
+// showHelp prints the help message.
+func showHelp(args *fuseArgs) {
+	executableName := sharefsName
+	if args.argc > 0 {
+		executableName = filepath.Base(C.GoString(*args.argv))
+	} else if name, err := os.Executable(); err == nil {
+		executableName = filepath.Base(name)
+	}
+
+	fmt.Printf("usage: %s [options] <mountpoint>\n\n", executableName)
+
+	fmt.Println("Filesystem options:")
+	fmt.Println("    -o source=/data/media  source directory to be mounted")
+	fmt.Println("    -o setuid=1023         uid to run as when starting as root")
+	fmt.Println("    -o setgid=1023         gid to run as when starting as root")
+
+	fmt.Println("\nFUSE options:")
+	C.fuse_cmdline_help()
+	C.fuse_lib_help(args)
+}
+
+// parseOpts parses fuse options via fuse_opt_parse.
+func parseOpts(args *fuseArgs, setup *setupState, log *log.Logger) (ok bool) {
+	var unsafeOpts struct {
+		// Pathname to writable source directory.
+		source *C.char
+
+		// Whether to create source directory as root.
+		mkdir C.int
+
+		// Decimal string representation of uid to set when running as root.
+		setuid *C.char
+		// Decimal string representation of gid to set when running as root.
+		setgid *C.char
+
+		// Decimal string representation of open file descriptor to read
+		// setupState from.
+		//
+		// This is an internal detail for containerisation and must not be
+		// specified directly.
+		setup *C.char
+	}
+
+	if C.fuse_opt_parse(args, unsafe.Pointer(&unsafeOpts), &[]C.struct_fuse_opt{
+		{templ: C.CString("source=%s"), offset: C.ulong(unsafe.Offsetof(unsafeOpts.source)), value: 0},
+		{templ: C.CString("mkdir"), offset: C.ulong(unsafe.Offsetof(unsafeOpts.mkdir)), value: 1},
+		{templ: C.CString("setuid=%s"), offset: C.ulong(unsafe.Offsetof(unsafeOpts.setuid)), value: 0},
+		{templ: C.CString("setgid=%s"), offset: C.ulong(unsafe.Offsetof(unsafeOpts.setgid)), value: 0},
+
+		{templ: C.CString("setup=%s"), offset: C.ulong(unsafe.Offsetof(unsafeOpts.setup)), value: 0},
+
+		C._FUSE_OPT_END(),
+	}[0], nil) == -1 {
+		return false
+	}
+
+	if unsafeOpts.source != nil {
+		defer C.free(unsafe.Pointer(unsafeOpts.source))
+	}
+	if unsafeOpts.setuid != nil {
+		defer C.free(unsafe.Pointer(unsafeOpts.setuid))
+	}
+	if unsafeOpts.setgid != nil {
+		defer C.free(unsafe.Pointer(unsafeOpts.setgid))
+	}
+
+	if unsafeOpts.setup != nil {
+		defer C.free(unsafe.Pointer(unsafeOpts.setup))
+
+		if v, err := strconv.Atoi(C.GoString(unsafeOpts.setup)); err != nil || v < 3 {
+			log.Println("invalid value for option setup")
+			return false
+		} else {
+			r := os.NewFile(uintptr(v), "setup")
+			defer func() {
+				if err = r.Close(); err != nil {
+					log.Println(err)
+				}
+			}()
+			if err = gob.NewDecoder(r).Decode(setup); err != nil {
+				log.Println(err)
+				return false
+			}
+		}
+		if setup.Fuse < 3 {
+			log.Println("invalid file descriptor", setup.Fuse)
+			return false
+		}
+		return true
+	}
+
+	if unsafeOpts.source == nil {
+		showHelp(args)
+		return false
+	} else if a, err := check.NewAbs(C.GoString(unsafeOpts.source)); err != nil {
+		log.Println(err)
+		return false
+	} else {
+		setup.Source = a
+	}
+	setup.mkdir = unsafeOpts.mkdir != 0
+
+	if unsafeOpts.setuid == nil {
+		setup.Setuid = -1
+	} else if v, err := strconv.Atoi(C.GoString(unsafeOpts.setuid)); err != nil || v <= 0 {
+		log.Println("invalid value for option setuid")
+		return false
+	} else {
+		setup.Setuid = v
+	}
+	if unsafeOpts.setgid == nil {
+		setup.Setgid = -1
+	} else if v, err := strconv.Atoi(C.GoString(unsafeOpts.setgid)); err != nil || v <= 0 {
+		log.Println("invalid value for option setgid")
+		return false
+	} else {
+		setup.Setgid = v
+	}
+
+	return true
+}
+
+// copyArgs returns a heap allocated copy of an argument slice in fuse_args
+// representation.
+func copyArgs(s ...string) fuseArgs {
+	if len(s) == 0 {
+		return fuseArgs{argc: 0, argv: nil, allocated: 0}
+	}
+	args := unsafe.Slice((**C.char)(C.malloc(C.size_t(uintptr(len(s))*unsafe.Sizeof(s[0])))), len(s))
+	for i, arg := range s {
+		args[i] = C.CString(arg)
+	}
+	return fuseArgs{argc: C.int(len(s)), argv: &args[0], allocated: 1}
+}
+
+// freeArgs frees the contents of argument list.
+func freeArgs(args *fuseArgs) { C.fuse_opt_free_args(args) }
+
+// unsafeAddArgument adds an argument to fuseArgs via fuse_opt_add_arg.
+//
+// The last byte of arg must be 0.
+func unsafeAddArgument(args *fuseArgs, arg string) {
+	C.fuse_opt_add_arg(args, (*C.char)(unsafe.Pointer(unsafe.StringData(arg))))
+}
+
+func _main(s ...string) (exitCode int) {
+	msg := message.New(log.Default())
+	container.TryArgv0(msg)
+	runtime.LockOSThread()
+
+	// don't mask creation mode, kernel already did that
+	syscall.Umask(0)
+
+	var pinner runtime.Pinner
+	defer pinner.Unpin()
+
+	args := copyArgs(s...)
+	defer freeArgs(&args)
+
+	// this causes the kernel to enforce access control based on struct stat
+	// populated by sharefs_getattr
+	unsafeAddArgument(&args, "-odefault_permissions\x00")
+
+	var priv C.struct_sharefs_private
+	pinner.Pin(&priv)
+	var setup setupState
+	priv.setup = C.uintptr_t(cgo.NewHandle(&setup))
+	defer destroySetup(unsafe.Pointer(&priv))
+
+	var opts C.struct_fuse_cmdline_opts
+	if C.fuse_parse_cmdline(&args, &opts) != 0 {
+		return 1
+	}
+	if opts.mountpoint != nil {
+		defer C.free(unsafe.Pointer(opts.mountpoint))
+	}
+
+	if opts.show_version != 0 {
+		fmt.Println("hakurei version", info.Version())
+		fmt.Println("FUSE library version", C.GoString(C.fuse_pkgversion()))
+		C.fuse_lowlevel_version()
+		return 0
+	}
+
+	if opts.show_help != 0 {
+		showHelp(&args)
+		return 0
+	} else if opts.mountpoint == nil {
+		log.Println("no mountpoint specified")
+		return 2
+	} else {
+		// hack to keep fuse_parse_cmdline happy in the container
+		mountpoint := C.GoString(opts.mountpoint)
+		pathnameArg := -1
+		for i, arg := range s {
+			if arg == mountpoint {
+				pathnameArg = i
+				break
+			}
+		}
+		if pathnameArg < 0 {
+			log.Println("mountpoint must be absolute")
+			return 2
+		}
+		s[pathnameArg] = container.Nonexistent
+	}
+
+	if !parseOpts(&args, &setup, msg.GetLogger()) {
+		return 1
+	}
+	asRoot := os.Geteuid() == 0
+
+	if asRoot {
+		if setup.Setuid <= 0 || setup.Setgid <= 0 {
+			log.Println("setuid and setgid must not be 0")
+			return 1
+		}
+
+		if setup.Fuse >= 3 {
+			log.Println("filesystem daemon must not run as root")
+			return 1
+		}
+
+		if setup.mkdir {
+			if err := os.MkdirAll(setup.Source.String(), 0700); err != nil {
+				if !errors.Is(err, os.ErrExist) {
+					log.Println(err)
+					return 1
+				}
+				// skip setup for existing source directory
+			} else if err = os.Chown(setup.Source.String(), setup.Setuid, setup.Setgid); err != nil {
+				log.Println(err)
+				return 1
+			}
+		}
+	} else if setup.Fuse < 3 && (setup.Setuid > 0 || setup.Setgid > 0) {
+		log.Println("setuid and setgid has no effect when not starting as root")
+		return 1
+	} else if setup.mkdir {
+		log.Println("mkdir has no effect when not starting as root")
+		return 1
+	}
+
+	op := C.struct_fuse_operations{
+		init:    closure(C.sharefs_init),
+		destroy: closure(C.sharefs_destroy),
+
+		// implemented in fuse-helper.c
+		getattr:  closure(C.sharefs_getattr),
+		readdir:  closure(C.sharefs_readdir),
+		mkdir:    closure(C.sharefs_mkdir),
+		unlink:   closure(C.sharefs_unlink),
+		rmdir:    closure(C.sharefs_rmdir),
+		rename:   closure(C.sharefs_rename),
+		truncate: closure(C.sharefs_truncate),
+		utimens:  closure(C.sharefs_utimens),
+		create:   closure(C.sharefs_create),
+		open:     closure(C.sharefs_open),
+		read:     closure(C.sharefs_read),
+		write:    closure(C.sharefs_write),
+		statfs:   closure(C.sharefs_statfs),
+		release:  closure(C.sharefs_release),
+		fsync:    closure(C.sharefs_fsync),
+	}
+
+	fuse := C.fuse_new_fn(&args, &op, C.size_t(unsafe.Sizeof(op)), unsafe.Pointer(&priv))
+	if fuse == nil {
+		return 3
+	}
+	defer C.fuse_destroy(fuse)
+	se := C.fuse_get_session(fuse)
+
+	if setup.Fuse < 3 {
+		// unconfined, set up mount point and container
+		if C.fuse_mount(fuse, opts.mountpoint) != 0 {
+			return 4
+		}
+		// unmounted by initial process
+		defer func() {
+			if exitCode == 5 {
+				C.fuse_unmount(fuse)
+			}
+		}()
+
+		if asRoot {
+			if err := syscall.Setresgid(setup.Setgid, setup.Setgid, setup.Setgid); err != nil {
+				log.Printf("cannot set gid: %v", err)
+				return 5
+			}
+			if err := syscall.Setgroups(nil); err != nil {
+				log.Printf("cannot set supplementary groups: %v", err)
+				return 5
+			}
+			if err := syscall.Setresuid(setup.Setuid, setup.Setuid, setup.Setuid); err != nil {
+				log.Printf("cannot set uid: %v", err)
+				return 5
+			}
+		}
+
+		msg.SwapVerbose(opts.debug != 0)
+		ctx := context.Background()
+		if opts.foreground != 0 {
+			c, cancel := signal.NotifyContext(ctx, syscall.SIGINT, syscall.SIGTERM)
+			defer cancel()
+			ctx = c
+		}
+		z := container.New(ctx, msg)
+		z.AllowOrphan = opts.foreground == 0
+		z.Env = os.Environ()
+
+		// keep fuse_parse_cmdline happy in the container
+		z.Tmpfs(check.MustAbs(container.Nonexistent), 1<<10, 0755)
+
+		z.Path = fhs.AbsProcSelfExe
+		z.Args = s
+		z.ForwardCancel = true
+		z.SeccompPresets |= std.PresetStrict
+		z.ParentPerm = 0700
+		z.Bind(setup.Source, setup.Source, std.BindWritable)
+		if !z.AllowOrphan {
+			z.WaitDelay = hst.WaitDelayMax
+			z.Stdin, z.Stdout, z.Stderr = os.Stdin, os.Stdout, os.Stderr
+		}
+		z.Bind(z.Path, z.Path, 0)
+		setup.Fuse = int(proc.ExtraFileSlice(
+			&z.ExtraFiles,
+			os.NewFile(uintptr(C.fuse_session_fd(se)), "fuse"),
+		))
+
+		var setupPipe [2]*os.File
+		if r, w, err := os.Pipe(); err != nil {
+			log.Println(err)
+			return 5
+		} else {
+			z.Args = append(z.Args, "-osetup="+strconv.Itoa(3+len(z.ExtraFiles)))
+			z.ExtraFiles = append(z.ExtraFiles, r)
+			setupPipe[0], setupPipe[1] = r, w
+		}
+
+		if err := z.Start(); err != nil {
+			if m, ok := message.GetMessage(err); ok {
+				log.Println(m)
+			} else {
+				log.Println(err)
+			}
+			return 5
+		}
+		if err := setupPipe[0].Close(); err != nil {
+			log.Println(err)
+		}
+		if err := z.Serve(); err != nil {
+			if m, ok := message.GetMessage(err); ok {
+				log.Println(m)
+			} else {
+				log.Println(err)
+			}
+			return 5
+		}
+
+		if err := gob.NewEncoder(setupPipe[1]).Encode(&setup); err != nil {
+			log.Println(err)
+			return 5
+		} else if err = setupPipe[1].Close(); err != nil {
+			log.Println(err)
+		}
+
+		if !z.AllowOrphan {
+			if err := z.Wait(); err != nil {
+				exitError, ok := errors.AsType[*exec.ExitError](err)
+				if !ok || exitError == nil {
+					log.Println(err)
+					return 5
+				}
+				switch code := exitError.ExitCode(); syscall.Signal(code & 0x7f) {
+				case syscall.SIGINT:
+				case syscall.SIGTERM:
+
+				default:
+					return code
+				}
+			}
+		}
+		return 0
+	} else { // confined
+		C.free(unsafe.Pointer(opts.mountpoint))
+		// must be heap allocated
+		opts.mountpoint = C.CString("/dev/fd/" + strconv.Itoa(setup.Fuse))
+
+		if err := os.Chdir("/"); err != nil {
+			log.Println(err)
+		}
+	}
+
+	if C.fuse_mount(fuse, opts.mountpoint) != 0 {
+		return 4
+	}
+	defer C.fuse_unmount(fuse)
+
+	if C.fuse_set_signal_handlers(se) != 0 {
+		return 6
+	}
+	defer C.fuse_remove_signal_handlers(se)
+
+	if opts.singlethread != 0 {
+		if C.fuse_loop(fuse) != 0 {
+			return 8
+		}
+	} else {
+		loopConfig := C.fuse_loop_cfg_create()
+		if loopConfig == nil {
+			return 7
+		}
+		defer C.fuse_loop_cfg_destroy(loopConfig)
+
+		C.fuse_loop_cfg_set_clone_fd(loopConfig, C.uint(opts.clone_fd))
+
+		C.fuse_loop_cfg_set_idle_threads(loopConfig, opts.max_idle_threads)
+		C.fuse_loop_cfg_set_max_threads(loopConfig, opts.max_threads)
+		if C.fuse_loop_mt(fuse, loopConfig) != 0 {
+			return 8
+		}
+	}
+
+	if setup.initFailed {
+		return 1
+	}
+	return 0
+}

cmd/sharefs/fuse_test.go

@@ -0,0 +1,113 @@
+package main
+
+import (
+	"bytes"
+	"log"
+	"reflect"
+	"testing"
+
+	"hakurei.app/check"
+)
+
+func TestParseOpts(t *testing.T) {
+	t.Parallel()
+
+	testCases := []struct {
+		name    string
+		args    []string
+		want    setupState
+		wantLog string
+		wantOk  bool
+	}{
+		{"zero length", []string{}, setupState{}, "", false},
+
+		{"not absolute", []string{"sharefs",
+			"-o", "source=nonexistent",
+			"-o", "setuid=1023",
+			"-o", "setgid=1023",
+		}, setupState{}, "sharefs: path \"nonexistent\" is not absolute\n", false},
+
+		{"not specified", []string{"sharefs",
+			"-o", "setuid=1023",
+			"-o", "setgid=1023",
+		}, setupState{}, "", false},
+
+		{"invalid setuid", []string{"sharefs",
+			"-o", "source=/proc/nonexistent",
+			"-o", "setuid=ff",
+			"-o", "setgid=1023",
+		}, setupState{
+			Source: check.MustAbs("/proc/nonexistent"),
+		}, "sharefs: invalid value for option setuid\n", false},
+
+		{"invalid setgid", []string{"sharefs",
+			"-o", "source=/proc/nonexistent",
+			"-o", "setuid=1023",
+			"-o", "setgid=ff",
+		}, setupState{
+			Source: check.MustAbs("/proc/nonexistent"),
+			Setuid: 1023,
+		}, "sharefs: invalid value for option setgid\n", false},
+
+		{"simple", []string{"sharefs",
+			"-o", "source=/proc/nonexistent",
+		}, setupState{
+			Source: check.MustAbs("/proc/nonexistent"),
+			Setuid: -1,
+			Setgid: -1,
+		}, "", true},
+
+		{"root", []string{"sharefs",
+			"-o", "source=/proc/nonexistent",
+			"-o", "setuid=1023",
+			"-o", "setgid=1023",
+		}, setupState{
+			Source: check.MustAbs("/proc/nonexistent"),
+			Setuid: 1023,
+			Setgid: 1023,
+		}, "", true},
+
+		{"setuid", []string{"sharefs",
+			"-o", "source=/proc/nonexistent",
+			"-o", "setuid=1023",
+		}, setupState{
+			Source: check.MustAbs("/proc/nonexistent"),
+			Setuid: 1023,
+			Setgid: -1,
+		}, "", true},
+
+		{"setgid", []string{"sharefs",
+			"-o", "source=/proc/nonexistent",
+			"-o", "setgid=1023",
+		}, setupState{
+			Source: check.MustAbs("/proc/nonexistent"),
+			Setuid: -1,
+			Setgid: 1023,
+		}, "", true},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			var (
+				got setupState
+				buf bytes.Buffer
+			)
+			args := copyArgs(tc.args...)
+			defer freeArgs(&args)
+			unsafeAddArgument(&args, "-odefault_permissions\x00")
+
+			if ok := parseOpts(&args, &got, log.New(&buf, "sharefs: ", 0)); ok != tc.wantOk {
+				t.Errorf("parseOpts: ok = %v, want %v", ok, tc.wantOk)
+			}
+
+			if !reflect.DeepEqual(&got, &tc.want) {
+				t.Errorf("parseOpts: setup = %#v, want %#v", got, tc.want)
+			}
+
+			if buf.String() != tc.wantLog {
+				t.Errorf("parseOpts: log =\n%s\nwant\n%s", buf.String(), tc.wantLog)
+			}
+		})
+	}
+}

cmd/sharefs/main.go

@@ -0,0 +1,38 @@
+// The sharefs FUSE filesystem is a permissionless shared filesystem.
+//
+// This filesystem is the primary means of file sharing between hakurei
+// application containers. It serves the same purpose in Rosa OS as /sdcard
+// does in AOSP.
+//
+// See help message for all available options.
+package main
+
+import (
+	"log"
+	"os"
+	"slices"
+)
+
+// sharefsName is the prefix used by log.std in the sharefs process.
+const sharefsName = "sharefs"
+
+// handleMountArgs returns an alternative, libfuse-compatible args slice for
+// args passed by mount -t fuse.sharefs [options] sharefs <mountpoint>.
+//
+// In this case, args always has a length of 5 with index 0 being what comes
+// after "fuse." in the filesystem type, 1 is the uninterpreted string passed
+// to mount (sharefsName is used as the magic string to enable this hack),
+// 2 is passed through to libfuse as mountpoint, and 3 is always "-o".
+func handleMountArgs(args []string) []string {
+	if len(args) == 5 && args[1] == sharefsName && args[3] == "-o" {
+		return []string{sharefsName, args[2], "-o", args[4]}
+	}
+	return slices.Clone(args)
+}
+
+func main() {
+	log.SetFlags(0)
+	log.SetPrefix(sharefsName + ": ")
+
+	os.Exit(_main(handleMountArgs(os.Args)...))
+}

cmd/sharefs/main_test.go

@@ -0,0 +1,29 @@
+package main
+
+import (
+	"slices"
+	"testing"
+)
+
+func TestHandleMountArgs(t *testing.T) {
+	t.Parallel()
+
+	testCases := []struct {
+		name string
+		args []string
+		want []string
+	}{
+		{"nil", nil, nil},
+		{"passthrough", []string{"sharefs", "-V"}, []string{"sharefs", "-V"}},
+		{"replace", []string{"/sbin/sharefs", "sharefs", "/sdcard", "-o", "rw"}, []string{"sharefs", "/sdcard", "-o", "rw"}},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			if got := handleMountArgs(tc.args); !slices.Equal(got, tc.want) {
+				t.Errorf("handleMountArgs: %q, want %q", got, tc.want)
+			}
+		})
+	}
+}

cmd/sharefs/test/configuration.nix

@@ -0,0 +1,44 @@
+{ pkgs, ... }:
+{
+  users.users = {
+    alice = {
+      isNormalUser = true;
+      description = "Alice Foobar";
+      password = "foobar";
+      uid = 1000;
+    };
+  };
+
+  home-manager.users.alice.home.stateVersion = "24.11";
+
+  # Automatically login on tty1 as a normal user:
+  services.getty.autologinUser = "alice";
+
+  environment = {
+    # For benchmarking sharefs:
+    systemPackages = [ pkgs.fsmark ];
+  };
+
+  virtualisation = {
+    # Hopefully reduces spurious test failures:
+    memorySize = if pkgs.stdenv.hostPlatform.is32bit then 2046 else 8192;
+
+    diskSize = 6 * 1024;
+
+    qemu.options = [
+      # Increase test performance:
+      "-smp 16"
+    ];
+  };
+
+  environment.hakurei = rec {
+    enable = true;
+    stateDir = "/var/lib/hakurei";
+    sharefs.source = "${stateDir}/sdcard";
+    users.alice = 0;
+
+    extraHomeConfig = {
+      home.stateVersion = "23.05";
+    };
+  };
+}

cmd/sharefs/test/default.nix

@@ -0,0 +1,44 @@
+{
+  testers,
+
+  system,
+  self,
+}:
+testers.nixosTest {
+  name = "sharefs";
+  nodes.machine =
+    { options, pkgs, ... }:
+    let
+      fhs =
+        let
+          hakurei = options.environment.hakurei.package.default;
+        in
+        pkgs.buildFHSEnv {
+          pname = "hakurei-fhs";
+          inherit (hakurei) version;
+          targetPkgs = _: hakurei.targetPkgs;
+          extraOutputsToInstall = [ "dev" ];
+          profile = ''
+            export PKG_CONFIG_PATH="/usr/share/pkgconfig:$PKG_CONFIG_PATH"
+          '';
+        };
+    in
+    {
+      environment.systemPackages = [
+        # For go tests:
+        (pkgs.writeShellScriptBin "sharefs-workload-hakurei-tests" ''
+          cp -r "${self.packages.${system}.hakurei.src}" "/sdcard/hakurei" && cd "/sdcard/hakurei"
+          ${fhs}/bin/hakurei-fhs -c 'ROSA_SKIP_BINFMT=1 CC="clang -O3 -Werror" go test ./...'
+        '')
+      ];
+
+      imports = [
+        ./configuration.nix
+
+        self.nixosModules.hakurei
+        self.inputs.home-manager.nixosModules.home-manager
+      ];
+    };
+
+  testScript = builtins.readFile ./test.py;
+}

cmd/sharefs/test/raceattr.go

@@ -0,0 +1,122 @@
+//go:build raceattr
+
+// The raceattr program reproduces vfs inode file attribute race.
+//
+// Even though libfuse high-level API presents the address of a struct stat
+// alongside struct fuse_context, file attributes are actually inherent to the
+// inode, instead of the specific call from userspace. The kernel implementation
+// in fs/fuse/xattr.c appears to make stale data in the inode (set by a previous
+// call) impossible or very unlikely to reach userspace via the stat family of
+// syscalls. However, when using default_permissions to have the VFS check
+// permissions, this race still happens, despite the resulting struct stat being
+// correct when overriding the check via capabilities otherwise.
+//
+// This program reproduces the failure, but because of its continuous nature, it
+// is provided independent of the vm integration test suite.
+package main
+
+import (
+	"context"
+	"flag"
+	"log"
+	"os"
+	"os/signal"
+	"runtime"
+	"sync"
+	"sync/atomic"
+	"syscall"
+)
+
+func newStatAs(
+	ctx context.Context, cancel context.CancelFunc,
+	n *atomic.Uint64, ok *atomic.Bool,
+	uid uint32, pathname string,
+	continuous bool,
+) func() {
+	return func() {
+		runtime.LockOSThread()
+		defer cancel()
+
+		if _, _, errno := syscall.Syscall(
+			syscall.SYS_SETUID, uintptr(uid),
+			0, 0,
+		); errno != 0 {
+			cancel()
+			log.Printf("cannot set uid to %d: %s", uid, errno)
+		}
+
+		var stat syscall.Stat_t
+		for {
+			if ctx.Err() != nil {
+				return
+			}
+
+			if err := syscall.Lstat(pathname, &stat); err != nil {
+				// SHAREFS_PERM_DIR not world executable, or
+				// SHAREFS_PERM_REG not world readable
+				if !continuous {
+					cancel()
+				}
+				ok.Store(true)
+				log.Printf("uid %d: %v", uid, err)
+			} else if stat.Uid != uid {
+				// appears to be unreachable
+				if !continuous {
+					cancel()
+				}
+				ok.Store(true)
+				log.Printf("got uid %d instead of %d", stat.Uid, uid)
+			}
+			n.Add(1)
+		}
+	}
+}
+
+func main() {
+	log.SetFlags(0)
+	log.SetPrefix("raceattr: ")
+
+	p := flag.String("target", "/sdcard/raceattr", "pathname of test file")
+	u0 := flag.Int("uid0", 1<<10-1, "first uid")
+	u1 := flag.Int("uid1", 1<<10-2, "second uid")
+	count := flag.Int("count", 1, "threads per uid")
+	continuous := flag.Bool("continuous", false, "keep running even after reproduce")
+	flag.Parse()
+
+	if os.Geteuid() != 0 {
+		log.Fatal("this program must run as root")
+	}
+
+	ctx, cancel := signal.NotifyContext(
+		context.Background(),
+		syscall.SIGINT,
+		syscall.SIGTERM,
+		syscall.SIGHUP,
+	)
+
+	if err := os.WriteFile(*p, nil, 0); err != nil {
+		log.Fatal(err)
+	}
+
+	var (
+		wg sync.WaitGroup
+
+		n  atomic.Uint64
+		ok atomic.Bool
+	)
+
+	if *count < 1 {
+		*count = 1
+	}
+	for range *count {
+		wg.Go(newStatAs(ctx, cancel, &n, &ok, uint32(*u0), *p, *continuous))
+		if *u1 >= 0 {
+			wg.Go(newStatAs(ctx, cancel, &n, &ok, uint32(*u1), *p, *continuous))
+		}
+	}
+
+	wg.Wait()
+	if !*continuous && ok.Load() {
+		log.Printf("reproduced after %d calls", n.Load())
+	}
+}

cmd/sharefs/test/test.py

@@ -0,0 +1,60 @@
+start_all()
+machine.wait_for_unit("multi-user.target")
+
+# To check sharefs version:
+print(machine.succeed("sharefs -V"))
+
+# Make sure sharefs started:
+machine.wait_for_unit("sdcard.mount")
+
+machine.succeed("mkdir /mnt")
+def check_bad_opts_output(opts, want, source="/etc", privileged=False):
+    output = machine.fail(("" if privileged else "sudo -u alice -i ") + f"sharefs -f -o source={source},{opts} /mnt 2>&1")
+    if output != want:
+        raise Exception(f"unexpected output: {output}")
+
+# Malformed setuid/setgid representation:
+check_bad_opts_output("setuid=ff", "sharefs: invalid value for option setuid\n")
+check_bad_opts_output("setgid=ff", "sharefs: invalid value for option setgid\n")
+
+# Bounds check for setuid/setgid:
+check_bad_opts_output("setuid=0", "sharefs: invalid value for option setuid\n")
+check_bad_opts_output("setgid=0", "sharefs: invalid value for option setgid\n")
+check_bad_opts_output("setuid=-1", "sharefs: invalid value for option setuid\n")
+check_bad_opts_output("setgid=-1", "sharefs: invalid value for option setgid\n")
+
+# Non-root setuid/setgid:
+check_bad_opts_output("setuid=1023", "sharefs: setuid and setgid has no effect when not starting as root\n")
+check_bad_opts_output("setgid=1023", "sharefs: setuid and setgid has no effect when not starting as root\n")
+check_bad_opts_output("setuid=1023,setgid=1023", "sharefs: setuid and setgid has no effect when not starting as root\n")
+check_bad_opts_output("mkdir", "sharefs: mkdir has no effect when not starting as root\n")
+
+# Starting as root without setuid/setgid:
+check_bad_opts_output("allow_other", "sharefs: setuid and setgid must not be 0\n", privileged=True)
+check_bad_opts_output("setuid=1023", "sharefs: setuid and setgid must not be 0\n", privileged=True)
+check_bad_opts_output("setgid=1023", "sharefs: setuid and setgid must not be 0\n", privileged=True)
+
+# Make sure nothing actually got mounted:
+machine.fail("umount /mnt")
+machine.succeed("rmdir /mnt")
+
+# Unprivileged mount/unmount:
+machine.succeed("sudo -u alice -i mkdir /home/alice/{sdcard,persistent}")
+machine.succeed("sudo -u alice -i sharefs -o source=/home/alice/persistent /home/alice/sdcard")
+machine.succeed("sudo -u alice -i touch /home/alice/sdcard/check")
+machine.succeed("sudo -u alice -i umount /home/alice/sdcard")
+machine.succeed("sudo -u alice -i rm /home/alice/persistent/check")
+machine.succeed("sudo -u alice -i rmdir /home/alice/{sdcard,persistent}")
+
+# Benchmark sharefs:
+machine.succeed("fs_mark -v -d /sdcard/fs_mark -l /tmp/fs_log.txt")
+machine.copy_from_vm("/tmp/fs_log.txt", "")
+
+# Check permissions:
+machine.succeed("sudo -u sharefs touch /var/lib/hakurei/sdcard/fs_mark/.check")
+machine.succeed("sudo -u sharefs rm /var/lib/hakurei/sdcard/fs_mark/.check")
+machine.succeed("sudo -u alice rm -rf /sdcard/fs_mark")
+machine.fail("ls /var/lib/hakurei/sdcard/fs_mark")
+
+# Run hakurei tests on sharefs:
+machine.succeed("sudo -u alice -i sharefs-workload-hakurei-tests")

command/builder_test.go

@@ -7,6 +7,7 @@ import (
 )
 
 func TestBuild(t *testing.T) {
+	t.Parallel()
 	c := command.New(nil, nil, "test", nil)
 	stubHandler := func([]string) error { panic("unreachable") }
 

command/parse.go

@@ -91,8 +91,8 @@ func (n *node) MustParse(arguments []string, handleError func(error)) {
 	case ErrEmptyTree:
 		os.Exit(1)
 	default:
-		var flagError FlagError
-		if !errors.As(err, &flagError) { // returned by HandlerFunc
+		flagError, ok := errors.AsType[FlagError](err)
+		if !ok { // returned by HandlerFunc
 			handleError(err)
 			os.Exit(1)
 		}

command/parse_test.go

@@ -14,6 +14,8 @@ import (
 )
 
 func TestParse(t *testing.T) {
+	t.Parallel()
+
 	testCases := []struct {
 		name      string
 		buildTree func(wout, wlog io.Writer) command.Command
@@ -251,6 +253,7 @@ Commands:
 	}
 	for _, tc := range testCases {
 		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
 			wout, wlog := new(bytes.Buffer), new(bytes.Buffer)
 			c := tc.buildTree(wout, wlog)
 

command/unreachable_test.go

@@ -6,15 +6,19 @@ import (
 )
 
 func TestParseUnreachable(t *testing.T) {
+	t.Parallel()
+
 	// top level bypasses name matching and recursive calls to Parse
 	// returns when encountering zero-length args
 	t.Run("zero-length args", func(t *testing.T) {
+		t.Parallel()
 		defer checkRecover(t, "Parse", "attempted to parse with zero length args")
 		_ = newNode(panicWriter{}, nil, " ", " ").Parse(nil)
 	})
 
 	// top level must not have siblings
 	t.Run("toplevel siblings", func(t *testing.T) {
+		t.Parallel()
 		defer checkRecover(t, "Parse", "invalid toplevel state")
 		n := newNode(panicWriter{}, nil, " ", "")
 		n.append(newNode(panicWriter{}, nil, "  ", " "))
@@ -23,6 +27,7 @@ func TestParseUnreachable(t *testing.T) {
 
 	// a node with descendents must not have a direct handler
 	t.Run("sub handle conflict", func(t *testing.T) {
+		t.Parallel()
 		defer checkRecover(t, "Parse", "invalid subcommand tree state")
 		n := newNode(panicWriter{}, nil, " ", " ")
 		n.adopt(newNode(panicWriter{}, nil, " ", " "))
@@ -32,6 +37,7 @@ func TestParseUnreachable(t *testing.T) {
 
 	// this would only happen if a node was matched twice
 	t.Run("parsed flag set", func(t *testing.T) {
+		t.Parallel()
 		defer checkRecover(t, "Parse", "invalid set state")
 		n := newNode(panicWriter{}, nil, " ", "")
 		set := flag.NewFlagSet("parsed", flag.ContinueOnError)

container/absolute.go

@@ -1,107 +0,0 @@
-package container
-
-import (
-	"encoding/json"
-	"errors"
-	"fmt"
-	"path"
-	"slices"
-	"strings"
-	"syscall"
-)
-
-// AbsoluteError is returned by [NewAbs] and holds the invalid pathname.
-type AbsoluteError struct {
-	Pathname string
-}
-
-func (e *AbsoluteError) Error() string { return fmt.Sprintf("path %q is not absolute", e.Pathname) }
-func (e *AbsoluteError) Is(target error) bool {
-	var ce *AbsoluteError
-	if !errors.As(target, &ce) {
-		return errors.Is(target, syscall.EINVAL)
-	}
-	return *e == *ce
-}
-
-// Absolute holds a pathname checked to be absolute.
-type Absolute struct {
-	pathname string
-}
-
-// isAbs wraps [path.IsAbs] in case additional checks are added in the future.
-func isAbs(pathname string) bool { return path.IsAbs(pathname) }
-
-func (a *Absolute) String() string {
-	if a.pathname == zeroString {
-		panic("attempted use of zero Absolute")
-	}
-	return a.pathname
-}
-
-func (a *Absolute) Is(v *Absolute) bool {
-	if a == nil && v == nil {
-		return true
-	}
-	return a != nil && v != nil &&
-		a.pathname != zeroString && v.pathname != zeroString &&
-		a.pathname == v.pathname
-}
-
-// NewAbs checks pathname and returns a new [Absolute] if pathname is absolute.
-func NewAbs(pathname string) (*Absolute, error) {
-	if !isAbs(pathname) {
-		return nil, &AbsoluteError{pathname}
-	}
-	return &Absolute{pathname}, nil
-}
-
-// MustAbs calls [NewAbs] and panics on error.
-func MustAbs(pathname string) *Absolute {
-	if a, err := NewAbs(pathname); err != nil {
-		panic(err.Error())
-	} else {
-		return a
-	}
-}
-
-// Append calls [path.Join] with [Absolute] as the first element.
-func (a *Absolute) Append(elem ...string) *Absolute {
-	return &Absolute{path.Join(append([]string{a.String()}, elem...)...)}
-}
-
-// Dir calls [path.Dir] with [Absolute] as its argument.
-func (a *Absolute) Dir() *Absolute { return &Absolute{path.Dir(a.String())} }
-
-func (a *Absolute) GobEncode() ([]byte, error) { return []byte(a.String()), nil }
-func (a *Absolute) GobDecode(data []byte) error {
-	pathname := string(data)
-	if !isAbs(pathname) {
-		return &AbsoluteError{pathname}
-	}
-	a.pathname = pathname
-	return nil
-}
-
-func (a *Absolute) MarshalJSON() ([]byte, error) { return json.Marshal(a.String()) }
-func (a *Absolute) UnmarshalJSON(data []byte) error {
-	var pathname string
-	if err := json.Unmarshal(data, &pathname); err != nil {
-		return err
-	}
-	if !isAbs(pathname) {
-		return &AbsoluteError{pathname}
-	}
-	a.pathname = pathname
-	return nil
-}
-
-// SortAbs calls [slices.SortFunc] for a slice of [Absolute].
-func SortAbs(x []*Absolute) {
-	slices.SortFunc(x, func(a, b *Absolute) int { return strings.Compare(a.String(), b.String()) })
-}
-
-// CompactAbs calls [slices.CompactFunc] for a slice of [Absolute].
-func CompactAbs(s []*Absolute) []*Absolute {
-	return slices.CompactFunc(s, func(a *Absolute, b *Absolute) bool { return a.String() == b.String() })
-}

container/autoetc.go

@@ -3,39 +3,43 @@ package container
 import (
 	"encoding/gob"
 	"fmt"
-	"io/fs"
+
+	"hakurei.app/check"
+	"hakurei.app/fhs"
 )
 
 func init() { gob.Register(new(AutoEtcOp)) }
 
-// Etc appends an [Op] that expands host /etc into a toplevel symlink mirror with /etc semantics.
-// This is not a generic setup op. It is implemented here to reduce ipc overhead.
-func (f *Ops) Etc(host *Absolute, prefix string) *Ops {
+// Etc is a helper for appending [AutoEtcOp] to [Ops].
+func (f *Ops) Etc(host *check.Absolute, prefix string) *Ops {
 	e := &AutoEtcOp{prefix}
-	f.Mkdir(AbsFHSEtc, 0755)
+	f.Mkdir(fhs.AbsEtc, 0755)
 	f.Bind(host, e.hostPath(), 0)
 	*f = append(*f, e)
 	return f
 }
 
+// AutoEtcOp expands host /etc into a toplevel symlink mirror with /etc semantics.
+//
+// This is not a generic setup op. It is implemented here to reduce ipc overhead.
 type AutoEtcOp struct{ Prefix string }
 
 func (e *AutoEtcOp) Valid() bool                                { return e != nil }
 func (e *AutoEtcOp) early(*setupState, syscallDispatcher) error { return nil }
 func (e *AutoEtcOp) apply(state *setupState, k syscallDispatcher) error {
 	if state.nonrepeatable&nrAutoEtc != 0 {
-		return msg.WrapErr(fs.ErrInvalid, "autoetc is not repeatable")
+		return OpRepeatError("autoetc")
 	}
 	state.nonrepeatable |= nrAutoEtc
 
-	const target = sysrootPath + FHSEtc
+	const target = sysrootPath + fhs.Etc
 	rel := e.hostRel() + "/"
 
 	if err := k.mkdirAll(target, 0755); err != nil {
-		return wrapErrSelf(err)
+		return err
 	}
 	if d, err := k.readdir(toSysroot(e.hostPath().String())); err != nil {
-		return wrapErrSelf(err)
+		return err
 	} else {
 		for _, ent := range d {
 			n := ent.Name()
@@ -43,13 +47,13 @@ func (e *AutoEtcOp) apply(state *setupState, k syscallDispatcher) error {
 			case ".host", "passwd", "group":
 
 			case "mtab":
-				if err = k.symlink(FHSProc+"mounts", target+n); err != nil {
-					return wrapErrSelf(err)
+				if err = k.symlink(fhs.Proc+"mounts", target+n); err != nil {
+					return err
 				}
 
 			default:
 				if err = k.symlink(rel+n, target+n); err != nil {
-					return wrapErrSelf(err)
+					return err
 				}
 			}
 		}
@@ -57,13 +61,14 @@ func (e *AutoEtcOp) apply(state *setupState, k syscallDispatcher) error {
 
 	return nil
 }
+func (e *AutoEtcOp) late(*setupState, syscallDispatcher) error { return nil }
 
-func (e *AutoEtcOp) hostPath() *Absolute { return AbsFHSEtc.Append(e.hostRel()) }
-func (e *AutoEtcOp) hostRel() string     { return ".host/" + e.Prefix }
+func (e *AutoEtcOp) hostPath() *check.Absolute { return fhs.AbsEtc.Append(e.hostRel()) }
+func (e *AutoEtcOp) hostRel() string           { return ".host/" + e.Prefix }
 
 func (e *AutoEtcOp) Is(op Op) bool {
 	ve, ok := op.(*AutoEtcOp)
 	return ok && e.Valid() && ve.Valid() && *e == *ve
 }
-func (*AutoEtcOp) prefix() string   { return "setting up" }
-func (e *AutoEtcOp) String() string { return fmt.Sprintf("auto etc %s", e.Prefix) }
+func (*AutoEtcOp) prefix() (string, bool) { return "setting up", true }
+func (e *AutoEtcOp) String() string       { return fmt.Sprintf("auto etc %s", e.Prefix) }

container/autoetc_test.go

@@ -2,14 +2,19 @@ package container
 
 import (
 	"errors"
-	"io/fs"
 	"os"
 	"testing"
+
+	"hakurei.app/check"
+	"hakurei.app/internal/stub"
 )
 
 func TestAutoEtcOp(t *testing.T) {
+	t.Parallel()
+
 	t.Run("nonrepeatable", func(t *testing.T) {
-		wantErr := msg.WrapErr(fs.ErrInvalid, "autoetc is not repeatable")
+		t.Parallel()
+		wantErr := OpRepeatError("autoetc")
 		if err := (&AutoEtcOp{Prefix: "81ceabb30d37bbdb3868004629cb84e9"}).apply(&setupState{nonrepeatable: nrAutoEtc}, nil); !errors.Is(err, wantErr) {
 			t.Errorf("apply: error = %v, want %v", err, wantErr)
 		}
@@ -18,22 +23,22 @@ func TestAutoEtcOp(t *testing.T) {
 	checkOpBehaviour(t, []opBehaviourTestCase{
 		{"mkdirAll", new(Params), &AutoEtcOp{
 			Prefix: "81ceabb30d37bbdb3868004629cb84e9",
-		}, nil, nil, []kexpect{
-			{"mkdirAll", expectArgs{"/sysroot/etc/", os.FileMode(0755)}, nil, errUnique},
-		}, wrapErrSelf(errUnique)},
+		}, nil, nil, []stub.Call{
+			call("mkdirAll", stub.ExpectArgs{"/sysroot/etc/", os.FileMode(0755)}, nil, stub.UniqueError(3)),
+		}, stub.UniqueError(3)},
 
 		{"readdir", new(Params), &AutoEtcOp{
 			Prefix: "81ceabb30d37bbdb3868004629cb84e9",
-		}, nil, nil, []kexpect{
-			{"mkdirAll", expectArgs{"/sysroot/etc/", os.FileMode(0755)}, nil, nil},
-			{"readdir", expectArgs{"/sysroot/etc/.host/81ceabb30d37bbdb3868004629cb84e9"}, stubDir(), errUnique},
-		}, wrapErrSelf(errUnique)},
+		}, nil, nil, []stub.Call{
+			call("mkdirAll", stub.ExpectArgs{"/sysroot/etc/", os.FileMode(0755)}, nil, nil),
+			call("readdir", stub.ExpectArgs{"/sysroot/etc/.host/81ceabb30d37bbdb3868004629cb84e9"}, stubDir(), stub.UniqueError(2)),
+		}, stub.UniqueError(2)},
 
 		{"symlink", new(Params), &AutoEtcOp{
 			Prefix: "81ceabb30d37bbdb3868004629cb84e9",
-		}, nil, nil, []kexpect{
-			{"mkdirAll", expectArgs{"/sysroot/etc/", os.FileMode(0755)}, nil, nil},
-			{"readdir", expectArgs{"/sysroot/etc/.host/81ceabb30d37bbdb3868004629cb84e9"}, stubDir(".host",
+		}, nil, nil, []stub.Call{
+			call("mkdirAll", stub.ExpectArgs{"/sysroot/etc/", os.FileMode(0755)}, nil, nil),
+			call("readdir", stub.ExpectArgs{"/sysroot/etc/.host/81ceabb30d37bbdb3868004629cb84e9"}, stubDir(".host",
 				"alsa", "bash_logout", "bashrc", "binfmt.d", "dbus-1", "default", "dhcpcd.exit-hook", "fonts",
 				"fstab", "fuse.conf", "group", "host.conf", "hostname", "hosts", "hsurc", "inputrc", "issue", "kbd",
 				"locale.conf", "login.defs", "lsb-release", "lvm", "machine-id", "man_db.conf", "mdadm.conf",
@@ -41,15 +46,15 @@ func TestAutoEtcOp(t *testing.T) {
 				"nsswitch.conf", "os-release", "pam", "pam.d", "passwd", "pipewire", "pki", "polkit-1", "profile",
 				"protocols", "resolv.conf", "resolvconf.conf", "rpc", "services", "set-environment", "shadow", "shells",
 				"ssh", "ssl", "static", "subgid", "subuid", "sudoers", "sway", "sysctl.d", "systemd", "terminfo",
-				"tmpfiles.d", "udev", "vconsole.conf", "X11", "xdg", "zoneinfo"), nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/alsa", "/sysroot/etc/alsa"}, nil, errUnique},
-		}, wrapErrSelf(errUnique)},
+				"tmpfiles.d", "udev", "vconsole.conf", "X11", "xdg", "zoneinfo"), nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/alsa", "/sysroot/etc/alsa"}, nil, stub.UniqueError(1)),
+		}, stub.UniqueError(1)},
 
 		{"symlink mtab", new(Params), &AutoEtcOp{
 			Prefix: "81ceabb30d37bbdb3868004629cb84e9",
-		}, nil, nil, []kexpect{
-			{"mkdirAll", expectArgs{"/sysroot/etc/", os.FileMode(0755)}, nil, nil},
-			{"readdir", expectArgs{"/sysroot/etc/.host/81ceabb30d37bbdb3868004629cb84e9"}, stubDir(".host",
+		}, nil, nil, []stub.Call{
+			call("mkdirAll", stub.ExpectArgs{"/sysroot/etc/", os.FileMode(0755)}, nil, nil),
+			call("readdir", stub.ExpectArgs{"/sysroot/etc/.host/81ceabb30d37bbdb3868004629cb84e9"}, stubDir(".host",
 				"alsa", "bash_logout", "bashrc", "binfmt.d", "dbus-1", "default", "dhcpcd.exit-hook", "fonts",
 				"fstab", "fuse.conf", "group", "host.conf", "hostname", "hosts", "hsurc", "inputrc", "issue", "kbd",
 				"locale.conf", "login.defs", "lsb-release", "lvm", "machine-id", "man_db.conf", "mdadm.conf",
@@ -57,41 +62,41 @@ func TestAutoEtcOp(t *testing.T) {
 				"nsswitch.conf", "os-release", "pam", "pam.d", "passwd", "pipewire", "pki", "polkit-1", "profile",
 				"protocols", "resolv.conf", "resolvconf.conf", "rpc", "services", "set-environment", "shadow", "shells",
 				"ssh", "ssl", "static", "subgid", "subuid", "sudoers", "sway", "sysctl.d", "systemd", "terminfo",
-				"tmpfiles.d", "udev", "vconsole.conf", "X11", "xdg", "zoneinfo"), nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/alsa", "/sysroot/etc/alsa"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/bash_logout", "/sysroot/etc/bash_logout"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/bashrc", "/sysroot/etc/bashrc"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/binfmt.d", "/sysroot/etc/binfmt.d"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/dbus-1", "/sysroot/etc/dbus-1"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/default", "/sysroot/etc/default"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/dhcpcd.exit-hook", "/sysroot/etc/dhcpcd.exit-hook"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fonts", "/sysroot/etc/fonts"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fstab", "/sysroot/etc/fstab"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fuse.conf", "/sysroot/etc/fuse.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/host.conf", "/sysroot/etc/host.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hostname", "/sysroot/etc/hostname"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hosts", "/sysroot/etc/hosts"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hsurc", "/sysroot/etc/hsurc"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/inputrc", "/sysroot/etc/inputrc"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/issue", "/sysroot/etc/issue"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/kbd", "/sysroot/etc/kbd"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/locale.conf", "/sysroot/etc/locale.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/login.defs", "/sysroot/etc/login.defs"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/lsb-release", "/sysroot/etc/lsb-release"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/lvm", "/sysroot/etc/lvm"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/machine-id", "/sysroot/etc/machine-id"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/man_db.conf", "/sysroot/etc/man_db.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/mdadm.conf", "/sysroot/etc/mdadm.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/modprobe.d", "/sysroot/etc/modprobe.d"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/modules-load.d", "/sysroot/etc/modules-load.d"}, nil, nil},
-			{"symlink", expectArgs{"/proc/mounts", "/sysroot/etc/mtab"}, nil, errUnique},
-		}, wrapErrSelf(errUnique)},
+				"tmpfiles.d", "udev", "vconsole.conf", "X11", "xdg", "zoneinfo"), nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/alsa", "/sysroot/etc/alsa"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/bash_logout", "/sysroot/etc/bash_logout"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/bashrc", "/sysroot/etc/bashrc"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/binfmt.d", "/sysroot/etc/binfmt.d"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/dbus-1", "/sysroot/etc/dbus-1"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/default", "/sysroot/etc/default"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/dhcpcd.exit-hook", "/sysroot/etc/dhcpcd.exit-hook"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fonts", "/sysroot/etc/fonts"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fstab", "/sysroot/etc/fstab"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fuse.conf", "/sysroot/etc/fuse.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/host.conf", "/sysroot/etc/host.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hostname", "/sysroot/etc/hostname"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hosts", "/sysroot/etc/hosts"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hsurc", "/sysroot/etc/hsurc"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/inputrc", "/sysroot/etc/inputrc"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/issue", "/sysroot/etc/issue"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/kbd", "/sysroot/etc/kbd"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/locale.conf", "/sysroot/etc/locale.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/login.defs", "/sysroot/etc/login.defs"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/lsb-release", "/sysroot/etc/lsb-release"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/lvm", "/sysroot/etc/lvm"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/machine-id", "/sysroot/etc/machine-id"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/man_db.conf", "/sysroot/etc/man_db.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/mdadm.conf", "/sysroot/etc/mdadm.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/modprobe.d", "/sysroot/etc/modprobe.d"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/modules-load.d", "/sysroot/etc/modules-load.d"}, nil, nil),
+			call("symlink", stub.ExpectArgs{"/proc/mounts", "/sysroot/etc/mtab"}, nil, stub.UniqueError(0)),
+		}, stub.UniqueError(0)},
 
 		{"success nested", new(Params), &AutoEtcOp{
 			Prefix: "81ceabb30d37bbdb3868004629cb84e9",
-		}, nil, nil, []kexpect{
-			{"mkdirAll", expectArgs{"/sysroot/etc/", os.FileMode(0755)}, nil, nil},
-			{"readdir", expectArgs{"/sysroot/etc/.host/81ceabb30d37bbdb3868004629cb84e9"}, stubDir(".host",
+		}, nil, nil, []stub.Call{
+			call("mkdirAll", stub.ExpectArgs{"/sysroot/etc/", os.FileMode(0755)}, nil, nil),
+			call("readdir", stub.ExpectArgs{"/sysroot/etc/.host/81ceabb30d37bbdb3868004629cb84e9"}, stubDir(".host",
 				"alsa", "bash_logout", "bashrc", "binfmt.d", "dbus-1", "default", "dhcpcd.exit-hook", "fonts",
 				"fstab", "fuse.conf", "group", "host.conf", "hostname", "hosts", "hsurc", "inputrc", "issue", "kbd",
 				"locale.conf", "login.defs", "lsb-release", "lvm", "machine-id", "man_db.conf", "mdadm.conf",
@@ -99,79 +104,79 @@ func TestAutoEtcOp(t *testing.T) {
 				"nsswitch.conf", "os-release", "pam", "pam.d", "passwd", "pipewire", "pki", "polkit-1", "profile",
 				"protocols", "resolv.conf", "resolvconf.conf", "rpc", "services", "set-environment", "shadow", "shells",
 				"ssh", "ssl", "static", "subgid", "subuid", "sudoers", "sway", "sysctl.d", "systemd", "terminfo",
-				"tmpfiles.d", "udev", "vconsole.conf", "X11", "xdg", "zoneinfo"), nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/alsa", "/sysroot/etc/alsa"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/bash_logout", "/sysroot/etc/bash_logout"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/bashrc", "/sysroot/etc/bashrc"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/binfmt.d", "/sysroot/etc/binfmt.d"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/dbus-1", "/sysroot/etc/dbus-1"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/default", "/sysroot/etc/default"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/dhcpcd.exit-hook", "/sysroot/etc/dhcpcd.exit-hook"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fonts", "/sysroot/etc/fonts"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fstab", "/sysroot/etc/fstab"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fuse.conf", "/sysroot/etc/fuse.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/host.conf", "/sysroot/etc/host.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hostname", "/sysroot/etc/hostname"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hosts", "/sysroot/etc/hosts"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hsurc", "/sysroot/etc/hsurc"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/inputrc", "/sysroot/etc/inputrc"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/issue", "/sysroot/etc/issue"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/kbd", "/sysroot/etc/kbd"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/locale.conf", "/sysroot/etc/locale.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/login.defs", "/sysroot/etc/login.defs"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/lsb-release", "/sysroot/etc/lsb-release"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/lvm", "/sysroot/etc/lvm"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/machine-id", "/sysroot/etc/machine-id"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/man_db.conf", "/sysroot/etc/man_db.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/mdadm.conf", "/sysroot/etc/mdadm.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/modprobe.d", "/sysroot/etc/modprobe.d"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/modules-load.d", "/sysroot/etc/modules-load.d"}, nil, nil},
-			{"symlink", expectArgs{"/proc/mounts", "/sysroot/etc/mtab"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nanorc", "/sysroot/etc/nanorc"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/netgroup", "/sysroot/etc/netgroup"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nix", "/sysroot/etc/nix"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nixos", "/sysroot/etc/nixos"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/NIXOS", "/sysroot/etc/NIXOS"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nscd.conf", "/sysroot/etc/nscd.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nsswitch.conf", "/sysroot/etc/nsswitch.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/os-release", "/sysroot/etc/os-release"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/pam", "/sysroot/etc/pam"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/pam.d", "/sysroot/etc/pam.d"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/pipewire", "/sysroot/etc/pipewire"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/pki", "/sysroot/etc/pki"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/polkit-1", "/sysroot/etc/polkit-1"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/profile", "/sysroot/etc/profile"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/protocols", "/sysroot/etc/protocols"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/resolv.conf", "/sysroot/etc/resolv.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/resolvconf.conf", "/sysroot/etc/resolvconf.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/rpc", "/sysroot/etc/rpc"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/services", "/sysroot/etc/services"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/set-environment", "/sysroot/etc/set-environment"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/shadow", "/sysroot/etc/shadow"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/shells", "/sysroot/etc/shells"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/ssh", "/sysroot/etc/ssh"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/ssl", "/sysroot/etc/ssl"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/static", "/sysroot/etc/static"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/subgid", "/sysroot/etc/subgid"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/subuid", "/sysroot/etc/subuid"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/sudoers", "/sysroot/etc/sudoers"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/sway", "/sysroot/etc/sway"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/sysctl.d", "/sysroot/etc/sysctl.d"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/systemd", "/sysroot/etc/systemd"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/terminfo", "/sysroot/etc/terminfo"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/tmpfiles.d", "/sysroot/etc/tmpfiles.d"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/udev", "/sysroot/etc/udev"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/vconsole.conf", "/sysroot/etc/vconsole.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/X11", "/sysroot/etc/X11"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/xdg", "/sysroot/etc/xdg"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/zoneinfo", "/sysroot/etc/zoneinfo"}, nil, nil},
+				"tmpfiles.d", "udev", "vconsole.conf", "X11", "xdg", "zoneinfo"), nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/alsa", "/sysroot/etc/alsa"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/bash_logout", "/sysroot/etc/bash_logout"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/bashrc", "/sysroot/etc/bashrc"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/binfmt.d", "/sysroot/etc/binfmt.d"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/dbus-1", "/sysroot/etc/dbus-1"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/default", "/sysroot/etc/default"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/dhcpcd.exit-hook", "/sysroot/etc/dhcpcd.exit-hook"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fonts", "/sysroot/etc/fonts"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fstab", "/sysroot/etc/fstab"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fuse.conf", "/sysroot/etc/fuse.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/host.conf", "/sysroot/etc/host.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hostname", "/sysroot/etc/hostname"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hosts", "/sysroot/etc/hosts"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hsurc", "/sysroot/etc/hsurc"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/inputrc", "/sysroot/etc/inputrc"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/issue", "/sysroot/etc/issue"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/kbd", "/sysroot/etc/kbd"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/locale.conf", "/sysroot/etc/locale.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/login.defs", "/sysroot/etc/login.defs"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/lsb-release", "/sysroot/etc/lsb-release"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/lvm", "/sysroot/etc/lvm"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/machine-id", "/sysroot/etc/machine-id"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/man_db.conf", "/sysroot/etc/man_db.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/mdadm.conf", "/sysroot/etc/mdadm.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/modprobe.d", "/sysroot/etc/modprobe.d"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/modules-load.d", "/sysroot/etc/modules-load.d"}, nil, nil),
+			call("symlink", stub.ExpectArgs{"/proc/mounts", "/sysroot/etc/mtab"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nanorc", "/sysroot/etc/nanorc"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/netgroup", "/sysroot/etc/netgroup"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nix", "/sysroot/etc/nix"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nixos", "/sysroot/etc/nixos"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/NIXOS", "/sysroot/etc/NIXOS"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nscd.conf", "/sysroot/etc/nscd.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nsswitch.conf", "/sysroot/etc/nsswitch.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/os-release", "/sysroot/etc/os-release"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/pam", "/sysroot/etc/pam"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/pam.d", "/sysroot/etc/pam.d"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/pipewire", "/sysroot/etc/pipewire"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/pki", "/sysroot/etc/pki"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/polkit-1", "/sysroot/etc/polkit-1"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/profile", "/sysroot/etc/profile"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/protocols", "/sysroot/etc/protocols"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/resolv.conf", "/sysroot/etc/resolv.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/resolvconf.conf", "/sysroot/etc/resolvconf.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/rpc", "/sysroot/etc/rpc"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/services", "/sysroot/etc/services"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/set-environment", "/sysroot/etc/set-environment"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/shadow", "/sysroot/etc/shadow"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/shells", "/sysroot/etc/shells"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/ssh", "/sysroot/etc/ssh"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/ssl", "/sysroot/etc/ssl"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/static", "/sysroot/etc/static"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/subgid", "/sysroot/etc/subgid"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/subuid", "/sysroot/etc/subuid"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/sudoers", "/sysroot/etc/sudoers"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/sway", "/sysroot/etc/sway"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/sysctl.d", "/sysroot/etc/sysctl.d"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/systemd", "/sysroot/etc/systemd"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/terminfo", "/sysroot/etc/terminfo"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/tmpfiles.d", "/sysroot/etc/tmpfiles.d"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/udev", "/sysroot/etc/udev"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/vconsole.conf", "/sysroot/etc/vconsole.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/X11", "/sysroot/etc/X11"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/xdg", "/sysroot/etc/xdg"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/zoneinfo", "/sysroot/etc/zoneinfo"}, nil, nil),
 		}, nil},
 
 		{"success", new(Params), &AutoEtcOp{
 			Prefix: "81ceabb30d37bbdb3868004629cb84e9",
-		}, nil, nil, []kexpect{
-			{"mkdirAll", expectArgs{"/sysroot/etc/", os.FileMode(0755)}, nil, nil},
-			{"readdir", expectArgs{"/sysroot/etc/.host/81ceabb30d37bbdb3868004629cb84e9"}, stubDir(
+		}, nil, nil, []stub.Call{
+			call("mkdirAll", stub.ExpectArgs{"/sysroot/etc/", os.FileMode(0755)}, nil, nil),
+			call("readdir", stub.ExpectArgs{"/sysroot/etc/.host/81ceabb30d37bbdb3868004629cb84e9"}, stubDir(
 				"alsa", "bash_logout", "bashrc", "binfmt.d", "dbus-1", "default", "dhcpcd.exit-hook", "fonts",
 				"fstab", "fuse.conf", "group", "host.conf", "hostname", "hosts", "hsurc", "inputrc", "issue", "kbd",
 				"locale.conf", "login.defs", "lsb-release", "lvm", "machine-id", "man_db.conf", "mdadm.conf",
@@ -179,72 +184,72 @@ func TestAutoEtcOp(t *testing.T) {
 				"nsswitch.conf", "os-release", "pam", "pam.d", "passwd", "pipewire", "pki", "polkit-1", "profile",
 				"protocols", "resolv.conf", "resolvconf.conf", "rpc", "services", "set-environment", "shadow", "shells",
 				"ssh", "ssl", "static", "subgid", "subuid", "sudoers", "sway", "sysctl.d", "systemd", "terminfo",
-				"tmpfiles.d", "udev", "vconsole.conf", "X11", "xdg", "zoneinfo"), nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/alsa", "/sysroot/etc/alsa"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/bash_logout", "/sysroot/etc/bash_logout"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/bashrc", "/sysroot/etc/bashrc"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/binfmt.d", "/sysroot/etc/binfmt.d"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/dbus-1", "/sysroot/etc/dbus-1"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/default", "/sysroot/etc/default"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/dhcpcd.exit-hook", "/sysroot/etc/dhcpcd.exit-hook"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fonts", "/sysroot/etc/fonts"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fstab", "/sysroot/etc/fstab"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fuse.conf", "/sysroot/etc/fuse.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/host.conf", "/sysroot/etc/host.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hostname", "/sysroot/etc/hostname"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hosts", "/sysroot/etc/hosts"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hsurc", "/sysroot/etc/hsurc"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/inputrc", "/sysroot/etc/inputrc"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/issue", "/sysroot/etc/issue"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/kbd", "/sysroot/etc/kbd"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/locale.conf", "/sysroot/etc/locale.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/login.defs", "/sysroot/etc/login.defs"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/lsb-release", "/sysroot/etc/lsb-release"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/lvm", "/sysroot/etc/lvm"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/machine-id", "/sysroot/etc/machine-id"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/man_db.conf", "/sysroot/etc/man_db.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/mdadm.conf", "/sysroot/etc/mdadm.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/modprobe.d", "/sysroot/etc/modprobe.d"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/modules-load.d", "/sysroot/etc/modules-load.d"}, nil, nil},
-			{"symlink", expectArgs{"/proc/mounts", "/sysroot/etc/mtab"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nanorc", "/sysroot/etc/nanorc"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/netgroup", "/sysroot/etc/netgroup"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nix", "/sysroot/etc/nix"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nixos", "/sysroot/etc/nixos"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/NIXOS", "/sysroot/etc/NIXOS"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nscd.conf", "/sysroot/etc/nscd.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nsswitch.conf", "/sysroot/etc/nsswitch.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/os-release", "/sysroot/etc/os-release"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/pam", "/sysroot/etc/pam"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/pam.d", "/sysroot/etc/pam.d"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/pipewire", "/sysroot/etc/pipewire"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/pki", "/sysroot/etc/pki"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/polkit-1", "/sysroot/etc/polkit-1"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/profile", "/sysroot/etc/profile"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/protocols", "/sysroot/etc/protocols"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/resolv.conf", "/sysroot/etc/resolv.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/resolvconf.conf", "/sysroot/etc/resolvconf.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/rpc", "/sysroot/etc/rpc"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/services", "/sysroot/etc/services"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/set-environment", "/sysroot/etc/set-environment"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/shadow", "/sysroot/etc/shadow"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/shells", "/sysroot/etc/shells"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/ssh", "/sysroot/etc/ssh"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/ssl", "/sysroot/etc/ssl"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/static", "/sysroot/etc/static"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/subgid", "/sysroot/etc/subgid"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/subuid", "/sysroot/etc/subuid"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/sudoers", "/sysroot/etc/sudoers"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/sway", "/sysroot/etc/sway"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/sysctl.d", "/sysroot/etc/sysctl.d"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/systemd", "/sysroot/etc/systemd"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/terminfo", "/sysroot/etc/terminfo"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/tmpfiles.d", "/sysroot/etc/tmpfiles.d"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/udev", "/sysroot/etc/udev"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/vconsole.conf", "/sysroot/etc/vconsole.conf"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/X11", "/sysroot/etc/X11"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/xdg", "/sysroot/etc/xdg"}, nil, nil},
-			{"symlink", expectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/zoneinfo", "/sysroot/etc/zoneinfo"}, nil, nil},
+				"tmpfiles.d", "udev", "vconsole.conf", "X11", "xdg", "zoneinfo"), nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/alsa", "/sysroot/etc/alsa"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/bash_logout", "/sysroot/etc/bash_logout"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/bashrc", "/sysroot/etc/bashrc"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/binfmt.d", "/sysroot/etc/binfmt.d"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/dbus-1", "/sysroot/etc/dbus-1"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/default", "/sysroot/etc/default"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/dhcpcd.exit-hook", "/sysroot/etc/dhcpcd.exit-hook"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fonts", "/sysroot/etc/fonts"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fstab", "/sysroot/etc/fstab"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/fuse.conf", "/sysroot/etc/fuse.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/host.conf", "/sysroot/etc/host.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hostname", "/sysroot/etc/hostname"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hosts", "/sysroot/etc/hosts"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/hsurc", "/sysroot/etc/hsurc"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/inputrc", "/sysroot/etc/inputrc"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/issue", "/sysroot/etc/issue"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/kbd", "/sysroot/etc/kbd"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/locale.conf", "/sysroot/etc/locale.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/login.defs", "/sysroot/etc/login.defs"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/lsb-release", "/sysroot/etc/lsb-release"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/lvm", "/sysroot/etc/lvm"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/machine-id", "/sysroot/etc/machine-id"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/man_db.conf", "/sysroot/etc/man_db.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/mdadm.conf", "/sysroot/etc/mdadm.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/modprobe.d", "/sysroot/etc/modprobe.d"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/modules-load.d", "/sysroot/etc/modules-load.d"}, nil, nil),
+			call("symlink", stub.ExpectArgs{"/proc/mounts", "/sysroot/etc/mtab"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nanorc", "/sysroot/etc/nanorc"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/netgroup", "/sysroot/etc/netgroup"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nix", "/sysroot/etc/nix"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nixos", "/sysroot/etc/nixos"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/NIXOS", "/sysroot/etc/NIXOS"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nscd.conf", "/sysroot/etc/nscd.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/nsswitch.conf", "/sysroot/etc/nsswitch.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/os-release", "/sysroot/etc/os-release"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/pam", "/sysroot/etc/pam"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/pam.d", "/sysroot/etc/pam.d"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/pipewire", "/sysroot/etc/pipewire"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/pki", "/sysroot/etc/pki"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/polkit-1", "/sysroot/etc/polkit-1"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/profile", "/sysroot/etc/profile"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/protocols", "/sysroot/etc/protocols"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/resolv.conf", "/sysroot/etc/resolv.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/resolvconf.conf", "/sysroot/etc/resolvconf.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/rpc", "/sysroot/etc/rpc"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/services", "/sysroot/etc/services"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/set-environment", "/sysroot/etc/set-environment"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/shadow", "/sysroot/etc/shadow"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/shells", "/sysroot/etc/shells"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/ssh", "/sysroot/etc/ssh"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/ssl", "/sysroot/etc/ssl"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/static", "/sysroot/etc/static"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/subgid", "/sysroot/etc/subgid"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/subuid", "/sysroot/etc/subuid"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/sudoers", "/sysroot/etc/sudoers"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/sway", "/sysroot/etc/sway"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/sysctl.d", "/sysroot/etc/sysctl.d"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/systemd", "/sysroot/etc/systemd"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/terminfo", "/sysroot/etc/terminfo"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/tmpfiles.d", "/sysroot/etc/tmpfiles.d"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/udev", "/sysroot/etc/udev"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/vconsole.conf", "/sysroot/etc/vconsole.conf"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/X11", "/sysroot/etc/X11"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/xdg", "/sysroot/etc/xdg"}, nil, nil),
+			call("symlink", stub.ExpectArgs{".host/81ceabb30d37bbdb3868004629cb84e9/zoneinfo", "/sysroot/etc/zoneinfo"}, nil, nil),
 		}, nil},
 	})
 
@@ -255,11 +260,11 @@ func TestAutoEtcOp(t *testing.T) {
 	})
 
 	checkOpsBuilder(t, []opsBuilderTestCase{
-		{"pd", new(Ops).Etc(MustAbs("/etc/"), "048090b6ed8f9ebb10e275ff5d8c0659"), Ops{
-			&MkdirOp{Path: MustAbs("/etc/"), Perm: 0755},
+		{"pd", new(Ops).Etc(check.MustAbs("/etc/"), "048090b6ed8f9ebb10e275ff5d8c0659"), Ops{
+			&MkdirOp{Path: check.MustAbs("/etc/"), Perm: 0755},
 			&BindMountOp{
-				Source: MustAbs("/etc/"),
-				Target: MustAbs("/etc/.host/048090b6ed8f9ebb10e275ff5d8c0659"),
+				Source: check.MustAbs("/etc/"),
+				Target: check.MustAbs("/etc/.host/048090b6ed8f9ebb10e275ff5d8c0659"),
 			},
 			&AutoEtcOp{Prefix: "048090b6ed8f9ebb10e275ff5d8c0659"},
 		}},
@@ -278,6 +283,7 @@ func TestAutoEtcOp(t *testing.T) {
 	})
 
 	t.Run("host path rel", func(t *testing.T) {
+		t.Parallel()
 		op := &AutoEtcOp{Prefix: "048090b6ed8f9ebb10e275ff5d8c0659"}
 		wantHostPath := "/etc/.host/048090b6ed8f9ebb10e275ff5d8c0659"
 		wantHostRel := ".host/048090b6ed8f9ebb10e275ff5d8c0659"

container/autoroot.go

@@ -3,42 +3,48 @@ package container
 import (
 	"encoding/gob"
 	"fmt"
-	"io/fs"
+
+	"hakurei.app/check"
+	"hakurei.app/fhs"
+	"hakurei.app/message"
 )
 
 func init() { gob.Register(new(AutoRootOp)) }
 
-// Root appends an [Op] that expands a directory into a toplevel bind mount mirror on container root.
-// This is not a generic setup op. It is implemented here to reduce ipc overhead.
-func (f *Ops) Root(host *Absolute, flags int) *Ops {
+// Root is a helper for appending [AutoRootOp] to [Ops].
+func (f *Ops) Root(host *check.Absolute, flags int) *Ops {
 	*f = append(*f, &AutoRootOp{host, flags, nil})
 	return f
 }
 
+// AutoRootOp expands a directory into a toplevel bind mount mirror on container root.
+//
+// This is not a generic setup op. It is implemented here to reduce ipc overhead.
 type AutoRootOp struct {
-	Host *Absolute
+	Host *check.Absolute
 	// passed through to bindMount
 	Flags int
 
 	// obtained during early;
 	// these wrap the underlying Op because BindMountOp is relatively complex,
 	// so duplicating that code would be unwise
-	resolved []Op
+	resolved []*BindMountOp
 }
 
 func (r *AutoRootOp) Valid() bool { return r != nil && r.Host != nil }
 
 func (r *AutoRootOp) early(state *setupState, k syscallDispatcher) error {
 	if d, err := k.readdir(r.Host.String()); err != nil {
-		return wrapErrSelf(err)
+		return err
 	} else {
-		r.resolved = make([]Op, 0, len(d))
+		r.resolved = make([]*BindMountOp, 0, len(d))
 		for _, ent := range d {
 			name := ent.Name()
-			if IsAutoRootBindable(name) {
+			if IsAutoRootBindable(state, name) {
+				// careful: the Valid method is skipped, make sure this is always valid
 				op := &BindMountOp{
 					Source: r.Host.Append(name),
-					Target: AbsFHSRoot.Append(name),
+					Target: fhs.AbsRoot.Append(name),
 					Flags:  r.Flags,
 				}
 				if err = op.early(state, k); err != nil {
@@ -53,18 +59,19 @@ func (r *AutoRootOp) early(state *setupState, k syscallDispatcher) error {
 
 func (r *AutoRootOp) apply(state *setupState, k syscallDispatcher) error {
 	if state.nonrepeatable&nrAutoRoot != 0 {
-		return msg.WrapErr(fs.ErrInvalid, "autoroot is not repeatable")
+		return OpRepeatError("autoroot")
 	}
 	state.nonrepeatable |= nrAutoRoot
 
 	for _, op := range r.resolved {
-		k.verbosef("%s %s", op.prefix(), op)
+		// these are exclusively BindMountOp, do not attempt to print identifying message
 		if err := op.apply(state, k); err != nil {
 			return err
 		}
 	}
 	return nil
 }
+func (r *AutoRootOp) late(*setupState, syscallDispatcher) error { return nil }
 
 func (r *AutoRootOp) Is(op Op) bool {
 	vr, ok := op.(*AutoRootOp)
@@ -72,13 +79,13 @@ func (r *AutoRootOp) Is(op Op) bool {
 		r.Host.Is(vr.Host) &&
 		r.Flags == vr.Flags
 }
-func (*AutoRootOp) prefix() string { return "setting up" }
+func (*AutoRootOp) prefix() (string, bool) { return "setting up", true }
 func (r *AutoRootOp) String() string {
 	return fmt.Sprintf("auto root %q flags %#x", r.Host, r.Flags)
 }
 
 // IsAutoRootBindable returns whether a dir entry name is selected for AutoRoot.
-func IsAutoRootBindable(name string) bool {
+func IsAutoRootBindable(msg message.Msg, name string) bool {
 	switch name {
 	case "proc", "dev", "tmp", "mnt", "etc":
 

container/autoroot_test.go

@@ -2,14 +2,19 @@ package container
 
 import (
 	"errors"
-	"io/fs"
 	"os"
 	"testing"
+
+	"hakurei.app/check"
+	"hakurei.app/container/std"
+	"hakurei.app/internal/stub"
+	"hakurei.app/message"
 )
 
 func TestAutoRootOp(t *testing.T) {
 	t.Run("nonrepeatable", func(t *testing.T) {
-		wantErr := msg.WrapErr(fs.ErrInvalid, "autoroot is not repeatable")
+		t.Parallel()
+		wantErr := OpRepeatError("autoroot")
 		if err := new(AutoRootOp).apply(&setupState{nonrepeatable: nrAutoRoot}, nil); !errors.Is(err, wantErr) {
 			t.Errorf("apply: error = %v, want %v", err, wantErr)
 		}
@@ -17,116 +22,115 @@ func TestAutoRootOp(t *testing.T) {
 
 	checkOpBehaviour(t, []opBehaviourTestCase{
 		{"readdir", &Params{ParentPerm: 0750}, &AutoRootOp{
-			Host:  MustAbs("/"),
-			Flags: BindWritable,
-		}, []kexpect{
-			{"readdir", expectArgs{"/"}, stubDir(), errUnique},
-		}, wrapErrSelf(errUnique), nil, nil},
+			Host:  check.MustAbs("/"),
+			Flags: std.BindWritable,
+		}, []stub.Call{
+			call("readdir", stub.ExpectArgs{"/"}, stubDir(), stub.UniqueError(2)),
+		}, stub.UniqueError(2), nil, nil},
 
 		{"early", &Params{ParentPerm: 0750}, &AutoRootOp{
-			Host:  MustAbs("/"),
-			Flags: BindWritable,
-		}, []kexpect{
-			{"readdir", expectArgs{"/"}, stubDir("bin", "dev", "etc", "home", "lib64",
-				"lost+found", "mnt", "nix", "proc", "root", "run", "srv", "sys", "tmp", "usr", "var"), nil},
-			{"evalSymlinks", expectArgs{"/bin"}, "", errUnique},
-		}, wrapErrSelf(errUnique), nil, nil},
+			Host:  check.MustAbs("/"),
+			Flags: std.BindWritable,
+		}, []stub.Call{
+			call("readdir", stub.ExpectArgs{"/"}, stubDir("bin", "dev", "etc", "home", "lib64",
+				"lost+found", "mnt", "nix", "proc", "root", "run", "srv", "sys", "tmp", "usr", "var"), nil),
+			call("evalSymlinks", stub.ExpectArgs{"/bin"}, "", stub.UniqueError(1)),
+		}, stub.UniqueError(1), nil, nil},
 
 		{"apply", &Params{ParentPerm: 0750}, &AutoRootOp{
-			Host:  MustAbs("/"),
-			Flags: BindWritable,
-		}, []kexpect{
-			{"readdir", expectArgs{"/"}, stubDir("bin", "dev", "etc", "home", "lib64",
-				"lost+found", "mnt", "nix", "proc", "root", "run", "srv", "sys", "tmp", "usr", "var"), nil},
-			{"evalSymlinks", expectArgs{"/bin"}, "/usr/bin", nil},
-			{"evalSymlinks", expectArgs{"/home"}, "/home", nil},
-			{"evalSymlinks", expectArgs{"/lib64"}, "/lib64", nil},
-			{"evalSymlinks", expectArgs{"/lost+found"}, "/lost+found", nil},
-			{"evalSymlinks", expectArgs{"/nix"}, "/nix", nil},
-			{"evalSymlinks", expectArgs{"/root"}, "/root", nil},
-			{"evalSymlinks", expectArgs{"/run"}, "/run", nil},
-			{"evalSymlinks", expectArgs{"/srv"}, "/srv", nil},
-			{"evalSymlinks", expectArgs{"/sys"}, "/sys", nil},
-			{"evalSymlinks", expectArgs{"/usr"}, "/usr", nil},
-			{"evalSymlinks", expectArgs{"/var"}, "/var", nil},
-		}, nil, []kexpect{
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/usr/bin"), MustAbs("/bin"), MustAbs("/bin"), BindWritable}}}, nil, nil},
-			{"stat", expectArgs{"/host/usr/bin"}, isDirFi(false), errUnique},
-		}, wrapErrSelf(errUnique)},
+			Host:  check.MustAbs("/"),
+			Flags: std.BindWritable,
+		}, []stub.Call{
+			call("readdir", stub.ExpectArgs{"/"}, stubDir("bin", "dev", "etc", "home", "lib64",
+				"lost+found", "mnt", "nix", "proc", "root", "run", "srv", "sys", "tmp", "usr", "var"), nil),
+			call("evalSymlinks", stub.ExpectArgs{"/bin"}, "/usr/bin", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/home"}, "/home", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/lib64"}, "/lib64", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/lost+found"}, "/lost+found", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/nix"}, "/nix", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/root"}, "/root", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/run"}, "/run", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/srv"}, "/srv", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/sys"}, "/sys", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/usr"}, "/usr", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/var"}, "/var", nil),
+		}, nil, []stub.Call{
+			call("stat", stub.ExpectArgs{"/host/usr/bin"}, isDirFi(false), stub.UniqueError(0)),
+		}, stub.UniqueError(0)},
 
 		{"success pd", &Params{ParentPerm: 0750}, &AutoRootOp{
-			Host:  MustAbs("/"),
-			Flags: BindWritable,
-		}, []kexpect{
-			{"readdir", expectArgs{"/"}, stubDir("bin", "dev", "etc", "home", "lib64",
-				"lost+found", "mnt", "nix", "proc", "root", "run", "srv", "sys", "tmp", "usr", "var"), nil},
-			{"evalSymlinks", expectArgs{"/bin"}, "/usr/bin", nil},
-			{"evalSymlinks", expectArgs{"/home"}, "/home", nil},
-			{"evalSymlinks", expectArgs{"/lib64"}, "/lib64", nil},
-			{"evalSymlinks", expectArgs{"/lost+found"}, "/lost+found", nil},
-			{"evalSymlinks", expectArgs{"/nix"}, "/nix", nil},
-			{"evalSymlinks", expectArgs{"/root"}, "/root", nil},
-			{"evalSymlinks", expectArgs{"/run"}, "/run", nil},
-			{"evalSymlinks", expectArgs{"/srv"}, "/srv", nil},
-			{"evalSymlinks", expectArgs{"/sys"}, "/sys", nil},
-			{"evalSymlinks", expectArgs{"/usr"}, "/usr", nil},
-			{"evalSymlinks", expectArgs{"/var"}, "/var", nil},
-		}, nil, []kexpect{
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/usr/bin"), MustAbs("/bin"), MustAbs("/bin"), BindWritable}}}, nil, nil}, {"stat", expectArgs{"/host/usr/bin"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/bin", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/usr/bin", "/sysroot/bin", uintptr(0x4004), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/home"), MustAbs("/home"), MustAbs("/home"), BindWritable}}}, nil, nil}, {"stat", expectArgs{"/host/home"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/home", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/home", "/sysroot/home", uintptr(0x4004), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/lib64"), MustAbs("/lib64"), MustAbs("/lib64"), BindWritable}}}, nil, nil}, {"stat", expectArgs{"/host/lib64"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/lib64", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/lib64", "/sysroot/lib64", uintptr(0x4004), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/lost+found"), MustAbs("/lost+found"), MustAbs("/lost+found"), BindWritable}}}, nil, nil}, {"stat", expectArgs{"/host/lost+found"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/lost+found", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/lost+found", "/sysroot/lost+found", uintptr(0x4004), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/nix"), MustAbs("/nix"), MustAbs("/nix"), BindWritable}}}, nil, nil}, {"stat", expectArgs{"/host/nix"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/nix", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/nix", "/sysroot/nix", uintptr(0x4004), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/root"), MustAbs("/root"), MustAbs("/root"), BindWritable}}}, nil, nil}, {"stat", expectArgs{"/host/root"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/root", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/root", "/sysroot/root", uintptr(0x4004), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/run"), MustAbs("/run"), MustAbs("/run"), BindWritable}}}, nil, nil}, {"stat", expectArgs{"/host/run"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/run", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/run", "/sysroot/run", uintptr(0x4004), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/srv"), MustAbs("/srv"), MustAbs("/srv"), BindWritable}}}, nil, nil}, {"stat", expectArgs{"/host/srv"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/srv", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/srv", "/sysroot/srv", uintptr(0x4004), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/sys"), MustAbs("/sys"), MustAbs("/sys"), BindWritable}}}, nil, nil}, {"stat", expectArgs{"/host/sys"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/sys", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/sys", "/sysroot/sys", uintptr(0x4004), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/usr"), MustAbs("/usr"), MustAbs("/usr"), BindWritable}}}, nil, nil}, {"stat", expectArgs{"/host/usr"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/usr", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/usr", "/sysroot/usr", uintptr(0x4004), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/var"), MustAbs("/var"), MustAbs("/var"), BindWritable}}}, nil, nil}, {"stat", expectArgs{"/host/var"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/var", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/var", "/sysroot/var", uintptr(0x4004), false}, nil, nil},
+			Host:  check.MustAbs("/"),
+			Flags: std.BindWritable,
+		}, []stub.Call{
+			call("readdir", stub.ExpectArgs{"/"}, stubDir("bin", "dev", "etc", "home", "lib64",
+				"lost+found", "mnt", "nix", "proc", "root", "run", "srv", "sys", "tmp", "usr", "var"), nil),
+			call("evalSymlinks", stub.ExpectArgs{"/bin"}, "/usr/bin", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/home"}, "/home", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/lib64"}, "/lib64", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/lost+found"}, "/lost+found", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/nix"}, "/nix", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/root"}, "/root", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/run"}, "/run", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/srv"}, "/srv", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/sys"}, "/sys", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/usr"}, "/usr", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/var"}, "/var", nil),
+		}, nil, []stub.Call{
+			call("stat", stub.ExpectArgs{"/host/usr/bin"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/bin", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q on %q flags %#x", []any{"/host/usr/bin", "/sysroot/bin", uintptr(0x4004)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/usr/bin", "/sysroot/bin", uintptr(0x4004), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/home"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/home", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q flags %#x", []any{"/sysroot/home", uintptr(0x4004)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/home", "/sysroot/home", uintptr(0x4004), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/lib64"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/lib64", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q flags %#x", []any{"/sysroot/lib64", uintptr(0x4004)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/lib64", "/sysroot/lib64", uintptr(0x4004), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/lost+found"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/lost+found", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q flags %#x", []any{"/sysroot/lost+found", uintptr(0x4004)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/lost+found", "/sysroot/lost+found", uintptr(0x4004), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/nix"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/nix", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q flags %#x", []any{"/sysroot/nix", uintptr(0x4004)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/nix", "/sysroot/nix", uintptr(0x4004), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/root"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/root", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q flags %#x", []any{"/sysroot/root", uintptr(0x4004)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/root", "/sysroot/root", uintptr(0x4004), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/run"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/run", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q flags %#x", []any{"/sysroot/run", uintptr(0x4004)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/run", "/sysroot/run", uintptr(0x4004), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/srv"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/srv", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q flags %#x", []any{"/sysroot/srv", uintptr(0x4004)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/srv", "/sysroot/srv", uintptr(0x4004), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/sys"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/sys", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q flags %#x", []any{"/sysroot/sys", uintptr(0x4004)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/sys", "/sysroot/sys", uintptr(0x4004), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/usr"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/usr", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q flags %#x", []any{"/sysroot/usr", uintptr(0x4004)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/usr", "/sysroot/usr", uintptr(0x4004), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/var"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/var", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q flags %#x", []any{"/sysroot/var", uintptr(0x4004)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/var", "/sysroot/var", uintptr(0x4004), false}, nil, nil),
 		}, nil},
 
 		{"success", &Params{ParentPerm: 0750}, &AutoRootOp{
-			Host: MustAbs("/var/lib/planterette/base/debian:f92c9052"),
-		}, []kexpect{
-			{"readdir", expectArgs{"/var/lib/planterette/base/debian:f92c9052"}, stubDir("bin", "dev", "etc", "home", "lib64",
-				"lost+found", "mnt", "nix", "proc", "root", "run", "srv", "sys", "tmp", "usr", "var"), nil},
-			{"evalSymlinks", expectArgs{"/var/lib/planterette/base/debian:f92c9052/bin"}, "/var/lib/planterette/base/debian:f92c9052/usr/bin", nil},
-			{"evalSymlinks", expectArgs{"/var/lib/planterette/base/debian:f92c9052/home"}, "/var/lib/planterette/base/debian:f92c9052/home", nil},
-			{"evalSymlinks", expectArgs{"/var/lib/planterette/base/debian:f92c9052/lib64"}, "/var/lib/planterette/base/debian:f92c9052/lib64", nil},
-			{"evalSymlinks", expectArgs{"/var/lib/planterette/base/debian:f92c9052/lost+found"}, "/var/lib/planterette/base/debian:f92c9052/lost+found", nil},
-			{"evalSymlinks", expectArgs{"/var/lib/planterette/base/debian:f92c9052/nix"}, "/var/lib/planterette/base/debian:f92c9052/nix", nil},
-			{"evalSymlinks", expectArgs{"/var/lib/planterette/base/debian:f92c9052/root"}, "/var/lib/planterette/base/debian:f92c9052/root", nil},
-			{"evalSymlinks", expectArgs{"/var/lib/planterette/base/debian:f92c9052/run"}, "/var/lib/planterette/base/debian:f92c9052/run", nil},
-			{"evalSymlinks", expectArgs{"/var/lib/planterette/base/debian:f92c9052/srv"}, "/var/lib/planterette/base/debian:f92c9052/srv", nil},
-			{"evalSymlinks", expectArgs{"/var/lib/planterette/base/debian:f92c9052/sys"}, "/var/lib/planterette/base/debian:f92c9052/sys", nil},
-			{"evalSymlinks", expectArgs{"/var/lib/planterette/base/debian:f92c9052/usr"}, "/var/lib/planterette/base/debian:f92c9052/usr", nil},
-			{"evalSymlinks", expectArgs{"/var/lib/planterette/base/debian:f92c9052/var"}, "/var/lib/planterette/base/debian:f92c9052/var", nil},
-		}, nil, []kexpect{
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/var/lib/planterette/base/debian:f92c9052/usr/bin"), MustAbs("/var/lib/planterette/base/debian:f92c9052/bin"), MustAbs("/bin"), 0}}}, nil, nil}, {"stat", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/usr/bin"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/bin", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/usr/bin", "/sysroot/bin", uintptr(0x4005), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/var/lib/planterette/base/debian:f92c9052/home"), MustAbs("/var/lib/planterette/base/debian:f92c9052/home"), MustAbs("/home"), 0}}}, nil, nil}, {"stat", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/home"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/home", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/home", "/sysroot/home", uintptr(0x4005), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/var/lib/planterette/base/debian:f92c9052/lib64"), MustAbs("/var/lib/planterette/base/debian:f92c9052/lib64"), MustAbs("/lib64"), 0}}}, nil, nil}, {"stat", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/lib64"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/lib64", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/lib64", "/sysroot/lib64", uintptr(0x4005), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/var/lib/planterette/base/debian:f92c9052/lost+found"), MustAbs("/var/lib/planterette/base/debian:f92c9052/lost+found"), MustAbs("/lost+found"), 0}}}, nil, nil}, {"stat", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/lost+found"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/lost+found", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/lost+found", "/sysroot/lost+found", uintptr(0x4005), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/var/lib/planterette/base/debian:f92c9052/nix"), MustAbs("/var/lib/planterette/base/debian:f92c9052/nix"), MustAbs("/nix"), 0}}}, nil, nil}, {"stat", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/nix"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/nix", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/nix", "/sysroot/nix", uintptr(0x4005), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/var/lib/planterette/base/debian:f92c9052/root"), MustAbs("/var/lib/planterette/base/debian:f92c9052/root"), MustAbs("/root"), 0}}}, nil, nil}, {"stat", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/root"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/root", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/root", "/sysroot/root", uintptr(0x4005), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/var/lib/planterette/base/debian:f92c9052/run"), MustAbs("/var/lib/planterette/base/debian:f92c9052/run"), MustAbs("/run"), 0}}}, nil, nil}, {"stat", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/run"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/run", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/run", "/sysroot/run", uintptr(0x4005), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/var/lib/planterette/base/debian:f92c9052/srv"), MustAbs("/var/lib/planterette/base/debian:f92c9052/srv"), MustAbs("/srv"), 0}}}, nil, nil}, {"stat", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/srv"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/srv", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/srv", "/sysroot/srv", uintptr(0x4005), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/var/lib/planterette/base/debian:f92c9052/sys"), MustAbs("/var/lib/planterette/base/debian:f92c9052/sys"), MustAbs("/sys"), 0}}}, nil, nil}, {"stat", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/sys"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/sys", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/sys", "/sysroot/sys", uintptr(0x4005), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/var/lib/planterette/base/debian:f92c9052/usr"), MustAbs("/var/lib/planterette/base/debian:f92c9052/usr"), MustAbs("/usr"), 0}}}, nil, nil}, {"stat", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/usr"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/usr", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/usr", "/sysroot/usr", uintptr(0x4005), false}, nil, nil},
-			{"verbosef", expectArgs{"%s %s", []any{"mounting", &BindMountOp{MustAbs("/var/lib/planterette/base/debian:f92c9052/var"), MustAbs("/var/lib/planterette/base/debian:f92c9052/var"), MustAbs("/var"), 0}}}, nil, nil}, {"stat", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/var"}, isDirFi(true), nil}, {"mkdirAll", expectArgs{"/sysroot/var", os.FileMode(0700)}, nil, nil}, {"bindMount", expectArgs{"/host/var/lib/planterette/base/debian:f92c9052/var", "/sysroot/var", uintptr(0x4005), false}, nil, nil},
+			Host: check.MustAbs("/var/lib/planterette/base/debian:f92c9052"),
+		}, []stub.Call{
+			call("readdir", stub.ExpectArgs{"/var/lib/planterette/base/debian:f92c9052"}, stubDir("bin", "dev", "etc", "home", "lib64",
+				"lost+found", "mnt", "nix", "proc", "root", "run", "srv", "sys", "tmp", "usr", "var"), nil),
+			call("evalSymlinks", stub.ExpectArgs{"/var/lib/planterette/base/debian:f92c9052/bin"}, "/var/lib/planterette/base/debian:f92c9052/usr/bin", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/var/lib/planterette/base/debian:f92c9052/home"}, "/var/lib/planterette/base/debian:f92c9052/home", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/var/lib/planterette/base/debian:f92c9052/lib64"}, "/var/lib/planterette/base/debian:f92c9052/lib64", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/var/lib/planterette/base/debian:f92c9052/lost+found"}, "/var/lib/planterette/base/debian:f92c9052/lost+found", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/var/lib/planterette/base/debian:f92c9052/nix"}, "/var/lib/planterette/base/debian:f92c9052/nix", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/var/lib/planterette/base/debian:f92c9052/root"}, "/var/lib/planterette/base/debian:f92c9052/root", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/var/lib/planterette/base/debian:f92c9052/run"}, "/var/lib/planterette/base/debian:f92c9052/run", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/var/lib/planterette/base/debian:f92c9052/srv"}, "/var/lib/planterette/base/debian:f92c9052/srv", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/var/lib/planterette/base/debian:f92c9052/sys"}, "/var/lib/planterette/base/debian:f92c9052/sys", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/var/lib/planterette/base/debian:f92c9052/usr"}, "/var/lib/planterette/base/debian:f92c9052/usr", nil),
+			call("evalSymlinks", stub.ExpectArgs{"/var/lib/planterette/base/debian:f92c9052/var"}, "/var/lib/planterette/base/debian:f92c9052/var", nil),
+		}, nil, []stub.Call{
+			call("stat", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/usr/bin"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/bin", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q on %q flags %#x", []any{"/host/var/lib/planterette/base/debian:f92c9052/usr/bin", "/sysroot/bin", uintptr(0x4005)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/usr/bin", "/sysroot/bin", uintptr(0x4005), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/home"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/home", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q on %q flags %#x", []any{"/host/var/lib/planterette/base/debian:f92c9052/home", "/sysroot/home", uintptr(0x4005)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/home", "/sysroot/home", uintptr(0x4005), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/lib64"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/lib64", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q on %q flags %#x", []any{"/host/var/lib/planterette/base/debian:f92c9052/lib64", "/sysroot/lib64", uintptr(0x4005)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/lib64", "/sysroot/lib64", uintptr(0x4005), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/lost+found"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/lost+found", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q on %q flags %#x", []any{"/host/var/lib/planterette/base/debian:f92c9052/lost+found", "/sysroot/lost+found", uintptr(0x4005)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/lost+found", "/sysroot/lost+found", uintptr(0x4005), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/nix"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/nix", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q on %q flags %#x", []any{"/host/var/lib/planterette/base/debian:f92c9052/nix", "/sysroot/nix", uintptr(0x4005)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/nix", "/sysroot/nix", uintptr(0x4005), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/root"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/root", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q on %q flags %#x", []any{"/host/var/lib/planterette/base/debian:f92c9052/root", "/sysroot/root", uintptr(0x4005)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/root", "/sysroot/root", uintptr(0x4005), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/run"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/run", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q on %q flags %#x", []any{"/host/var/lib/planterette/base/debian:f92c9052/run", "/sysroot/run", uintptr(0x4005)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/run", "/sysroot/run", uintptr(0x4005), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/srv"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/srv", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q on %q flags %#x", []any{"/host/var/lib/planterette/base/debian:f92c9052/srv", "/sysroot/srv", uintptr(0x4005)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/srv", "/sysroot/srv", uintptr(0x4005), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/sys"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/sys", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q on %q flags %#x", []any{"/host/var/lib/planterette/base/debian:f92c9052/sys", "/sysroot/sys", uintptr(0x4005)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/sys", "/sysroot/sys", uintptr(0x4005), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/usr"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/usr", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q on %q flags %#x", []any{"/host/var/lib/planterette/base/debian:f92c9052/usr", "/sysroot/usr", uintptr(0x4005)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/usr", "/sysroot/usr", uintptr(0x4005), false}, nil, nil),
+			call("stat", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/var"}, isDirFi(true), nil), call("mkdirAll", stub.ExpectArgs{"/sysroot/var", os.FileMode(0700)}, nil, nil), call("verbosef", stub.ExpectArgs{"mounting %q on %q flags %#x", []any{"/host/var/lib/planterette/base/debian:f92c9052/var", "/sysroot/var", uintptr(0x4005)}}, nil, nil), call("bindMount", stub.ExpectArgs{"/host/var/lib/planterette/base/debian:f92c9052/var", "/sysroot/var", uintptr(0x4005), false}, nil, nil),
 		}, nil},
 	})
 
 	checkOpsValid(t, []opValidTestCase{
 		{"nil", (*AutoRootOp)(nil), false},
 		{"zero", new(AutoRootOp), false},
-		{"valid", &AutoRootOp{Host: MustAbs("/")}, true},
+		{"valid", &AutoRootOp{Host: check.MustAbs("/")}, true},
 	})
 
 	checkOpsBuilder(t, []opsBuilderTestCase{
-		{"pd", new(Ops).Root(MustAbs("/"), BindWritable), Ops{
+		{"pd", new(Ops).Root(check.MustAbs("/"), std.BindWritable), Ops{
 			&AutoRootOp{
-				Host:  MustAbs("/"),
-				Flags: BindWritable,
+				Host:  check.MustAbs("/"),
+				Flags: std.BindWritable,
 			},
 		}},
 	})
@@ -135,64 +139,74 @@ func TestAutoRootOp(t *testing.T) {
 		{"zero", new(AutoRootOp), new(AutoRootOp), false},
 
 		{"internal ne", &AutoRootOp{
-			Host:  MustAbs("/"),
-			Flags: BindWritable,
+			Host:  check.MustAbs("/"),
+			Flags: std.BindWritable,
 		}, &AutoRootOp{
-			Host:     MustAbs("/"),
-			Flags:    BindWritable,
-			resolved: []Op{new(BindMountOp)},
+			Host:     check.MustAbs("/"),
+			Flags:    std.BindWritable,
+			resolved: []*BindMountOp{new(BindMountOp)},
 		}, true},
 
 		{"flags differs", &AutoRootOp{
-			Host:  MustAbs("/"),
-			Flags: BindWritable | BindDevice,
+			Host:  check.MustAbs("/"),
+			Flags: std.BindWritable | std.BindDevice,
 		}, &AutoRootOp{
-			Host:  MustAbs("/"),
-			Flags: BindWritable,
+			Host:  check.MustAbs("/"),
+			Flags: std.BindWritable,
 		}, false},
 
 		{"host differs", &AutoRootOp{
-			Host:  MustAbs("/tmp/"),
-			Flags: BindWritable,
+			Host:  check.MustAbs("/tmp/"),
+			Flags: std.BindWritable,
 		}, &AutoRootOp{
-			Host:  MustAbs("/"),
-			Flags: BindWritable,
+			Host:  check.MustAbs("/"),
+			Flags: std.BindWritable,
 		}, false},
 
 		{"equals", &AutoRootOp{
-			Host:  MustAbs("/"),
-			Flags: BindWritable,
+			Host:  check.MustAbs("/"),
+			Flags: std.BindWritable,
 		}, &AutoRootOp{
-			Host:  MustAbs("/"),
-			Flags: BindWritable,
+			Host:  check.MustAbs("/"),
+			Flags: std.BindWritable,
 		}, true},
 	})
 
 	checkOpMeta(t, []opMetaTestCase{
 		{"root", &AutoRootOp{
-			Host:  MustAbs("/"),
-			Flags: BindWritable,
+			Host:  check.MustAbs("/"),
+			Flags: std.BindWritable,
 		}, "setting up", `auto root "/" flags 0x2`},
 	})
 }
 
 func TestIsAutoRootBindable(t *testing.T) {
+	t.Parallel()
+
 	testCases := []struct {
 		name string
 		want bool
+		log  bool
 	}{
-		{"proc", false},
-		{"dev", false},
-		{"tmp", false},
-		{"mnt", false},
-		{"etc", false},
-		{"", false},
+		{"proc", false, false},
+		{"dev", false, false},
+		{"tmp", false, false},
+		{"mnt", false, false},
+		{"etc", false, false},
+		{"", false, true},
 
-		{"var", true},
+		{"var", true, false},
 	}
 	for _, tc := range testCases {
 		t.Run(tc.name, func(t *testing.T) {
-			if got := IsAutoRootBindable(tc.name); got != tc.want {
+			t.Parallel()
+			var msg message.Msg
+			if tc.log {
+				msg = &kstub{nil, nil, stub.New(t, func(s *stub.Stub[syscallDispatcher]) syscallDispatcher { panic("unreachable") }, stub.Expect{Calls: []stub.Call{
+					call("verbose", stub.ExpectArgs{[]any{"got unexpected root entry"}}, nil, nil),
+				}})}
+			}
+			if got := IsAutoRootBindable(msg, tc.name); got != tc.want {
 				t.Errorf("IsAutoRootBindable: %v, want %v", got, tc.want)
 			}
 		})

container/binfmt.go

@@ -0,0 +1,46 @@
+package container
+
+import (
+	"strings"
+	"unsafe"
+
+	"hakurei.app/check"
+)
+
+// escapeBinfmt escapes magic/mask sequences in a [BinfmtEntry].
+func escapeBinfmt(buf *strings.Builder, s string) string {
+	const lowerhex = "0123456789abcdef"
+
+	buf.Reset()
+	for _, c := range unsafe.Slice(unsafe.StringData(s), len(s)) {
+		switch c {
+		case 0, '\\', ':':
+			buf.WriteString(`\x`)
+			buf.WriteByte(lowerhex[c>>4])
+			buf.WriteByte(lowerhex[c&0xf])
+
+		default:
+			buf.WriteByte(c)
+		}
+	}
+	return buf.String()
+}
+
+// BinfmtEntry is an entry to be registered by the init process.
+type BinfmtEntry struct {
+	// The offset of the magic/mask in the file, counted in bytes.
+	Offset byte
+	// The byte sequence binfmt_misc is matching for.
+	Magic string
+	// An (optional, defaults to all 0xff) mask.
+	Mask string
+	// The program that should be invoked with the binary as first argument.
+	Interpreter *check.Absolute
+}
+
+// Valid returns whether e can be registered into the kernel.
+func (e *BinfmtEntry) Valid() bool {
+	return e != nil &&
+		int(e.Offset)+max(len(e.Magic), len(e.Mask)) < 128 &&
+		e.Interpreter != nil && len(e.Interpreter.String()) < 128
+}

container/binfmt_test.go

@@ -0,0 +1,62 @@
+package container
+
+import (
+	"strings"
+	"testing"
+
+	"hakurei.app/fhs"
+)
+
+func TestEscapeBinfmt(t *testing.T) {
+	t.Parallel()
+
+	testCases := []struct {
+		name  string
+		magic string
+		want  string
+	}{
+		{"packed DOS applications", "\x0eDEX", "\x0eDEX"},
+
+		{"riscv64 magic",
+			"\x7fELF\x02\x01\x01\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02\x00\xf3\x00",
+			"\x7fELF\x02\x01\x01\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\x02\\x00\xf3\\x00"},
+		{"riscv64 mask",
+			"\xff\xff\xff\xff\xff\xff\xff\x00\xff\xff\xff\xff\xff\xff\xff\xff\xfe\xff\xff\xff",
+			"\xff\xff\xff\xff\xff\xff\xff\\x00\xff\xff\xff\xff\xff\xff\xff\xff\xfe\xff\xff\xff"},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			got := escapeBinfmt(new(strings.Builder), tc.magic)
+			if got != tc.want {
+				t.Errorf("escapeBinfmt: %q, want %q", got, tc.want)
+			}
+		})
+	}
+}
+
+func TestBinfmtEntry(t *testing.T) {
+	t.Parallel()
+
+	testCases := []struct {
+		name  string
+		e     BinfmtEntry
+		valid bool
+	}{
+		{"zero", BinfmtEntry{}, false},
+		{"large offset", BinfmtEntry{Offset: 128}, false},
+		{"long magic", BinfmtEntry{Magic: strings.Repeat("\x00", 128)}, false},
+		{"long mask", BinfmtEntry{Mask: strings.Repeat("\x00", 128)}, false},
+		{"valid", BinfmtEntry{Interpreter: fhs.AbsRoot}, true},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			if tc.e.Valid() != tc.valid {
+				t.Errorf("Valid: %v", !tc.valid)
+			}
+		})
+	}
+}

container/capability.go

@@ -3,6 +3,8 @@ package container
 import (
 	"syscall"
 	"unsafe"
+
+	"hakurei.app/ext"
 )
 
 const (
@@ -14,7 +16,9 @@ const (
 
 	CAP_SYS_ADMIN    = 0x15
 	CAP_SETPCAP      = 0x8
+	CAP_NET_ADMIN    = 0xc
 	CAP_DAC_OVERRIDE = 0x1
+	CAP_SETFCAP      = 0x1f
 )
 
 type (
@@ -50,40 +54,15 @@ func capset(hdrp *capHeader, datap *[2]capData) error {
 
 // capBoundingSetDrop drops a capability from the calling thread's capability bounding set.
 func capBoundingSetDrop(cap uintptr) error {
-	r, _, errno := syscall.Syscall(
-		syscall.SYS_PRCTL,
-		syscall.PR_CAPBSET_DROP,
-		cap, 0,
-	)
-	if r != 0 {
-		return errno
-	}
-	return nil
+	return ext.Prctl(syscall.PR_CAPBSET_DROP, cap, 0)
 }
 
 // capAmbientClearAll clears the ambient capability set of the calling thread.
 func capAmbientClearAll() error {
-	r, _, errno := syscall.Syscall(
-		syscall.SYS_PRCTL,
-		PR_CAP_AMBIENT,
-		PR_CAP_AMBIENT_CLEAR_ALL, 0,
-	)
-	if r != 0 {
-		return errno
-	}
-	return nil
+	return ext.Prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_CLEAR_ALL, 0)
 }
 
 // capAmbientRaise adds to the ambient capability set of the calling thread.
 func capAmbientRaise(cap uintptr) error {
-	r, _, errno := syscall.Syscall(
-		syscall.SYS_PRCTL,
-		PR_CAP_AMBIENT,
-		PR_CAP_AMBIENT_RAISE,
-		cap,
-	)
-	if r != 0 {
-		return errno
-	}
-	return nil
+	return ext.Prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_RAISE, cap)
 }

container/capability_test.go

@@ -3,6 +3,8 @@ package container
 import "testing"
 
 func TestCapToIndex(t *testing.T) {
+	t.Parallel()
+
 	testCases := []struct {
 		name string
 		cap  uintptr
@@ -14,6 +16,7 @@ func TestCapToIndex(t *testing.T) {
 	}
 	for _, tc := range testCases {
 		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
 			if got := capToIndex(tc.cap); got != tc.want {
 				t.Errorf("capToIndex: %#x, want %#x", got, tc.want)
 			}
@@ -22,6 +25,8 @@ func TestCapToIndex(t *testing.T) {
 }
 
 func TestCapToMask(t *testing.T) {
+	t.Parallel()
+
 	testCases := []struct {
 		name string
 		cap  uintptr
@@ -33,6 +38,7 @@ func TestCapToMask(t *testing.T) {
 	}
 	for _, tc := range testCases {
 		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
 			if got := capToMask(tc.cap); got != tc.want {
 				t.Errorf("capToMask: %#x, want %#x", got, tc.want)
 			}

container/container.go

@@ -1,4 +1,5 @@
-// Package container implements unprivileged Linux containers with built-in support for syscall filtering.
+// Package container implements unprivileged Linux containers with built-in
+// support for syscall filtering.
 package container
 
 import (
@@ -11,77 +12,108 @@ import (
 	"os/exec"
 	"runtime"
 	"strconv"
+	"sync"
 	. "syscall"
 	"time"
 
+	"hakurei.app/check"
 	"hakurei.app/container/seccomp"
+	"hakurei.app/container/std"
+	"hakurei.app/ext"
+	"hakurei.app/fhs"
+	"hakurei.app/internal/landlock"
+	"hakurei.app/message"
 )
 
 const (
 	// CancelSignal is the signal expected by container init on context cancel.
 	// A custom [Container.Cancel] function must eventually deliver this signal.
-	CancelSignal = SIGTERM
+	CancelSignal = SIGUSR2
 )
 
 type (
 	// Container represents a container environment being prepared or run.
 	// None of [Container] methods are safe for concurrent use.
 	Container struct {
+		// Whether the container init should stay alive after its parent terminates.
+		AllowOrphan bool
+		// Whether to set SchedPolicy and SchedPriority via sched_setscheduler(2).
+		SetScheduler bool
+		// Scheduling policy to set via sched_setscheduler(2).
+		SchedPolicy ext.SchedPolicy
+		// Scheduling priority to set via sched_setscheduler(2). The zero value
+		// implies the minimum value supported by the current SchedPolicy.
+		SchedPriority ext.Int
 		// Cgroup fd, nil to disable.
 		Cgroup *int
-		// ExtraFiles passed through to initial process in the container,
-		// with behaviour identical to its [exec.Cmd] counterpart.
+		// ExtraFiles passed through to initial process in the container, with
+		// behaviour identical to its [exec.Cmd] counterpart.
 		ExtraFiles []*os.File
 
-		// param encoder for shim and init
-		setup *gob.Encoder
-		// cancels cmd
+		// Write end of a pipe connected to the init to deliver [Params].
+		setup [2]*os.File
+		// Cancels the context passed to the underlying cmd.
 		cancel context.CancelFunc
-		// closed after Wait returns
+		// Closed after Wait returns. Keeps the spawning thread alive.
 		wait chan struct{}
 
 		Stdin  io.Reader
 		Stdout io.Writer
 		Stderr io.Writer
 
-		Cancel    func(cmd *exec.Cmd) error
+		// Custom cancellation behaviour for the underlying [exec.Cmd]. Must
+		// deliver [CancelSignal] before returning.
+		Cancel func(cmd *exec.Cmd) error
+		// Copied to the underlying [exec.Cmd].
 		WaitDelay time.Duration
 
+		// Suppress verbose output of init.
+		Quiet bool
+
 		cmd *exec.Cmd
 		ctx context.Context
+		msg message.Msg
 		Params
 	}
 
 	// Params holds container configuration and is safe to serialise.
 	Params struct {
 		// Working directory in the container.
-		Dir *Absolute
+		Dir *check.Absolute
 		// Initial process environment.
 		Env []string
 		// Pathname of initial process in the container.
-		Path *Absolute
+		Path *check.Absolute
 		// Initial process argv.
 		Args []string
 		// Deliver SIGINT to the initial process on context cancellation.
 		ForwardCancel bool
-		// time to wait for linger processes after death of initial process
+		// Time to wait for processes lingering after the initial process terminates.
 		AdoptWaitDelay time.Duration
 
+		// Map uid/gid 0 in the init process. Requires [FstypeProc] attached to
+		// [fhs.Proc] in the container filesystem.
+		InitAsRoot bool
 		// Mapped Uid in user namespace.
 		Uid int
 		// Mapped Gid in user namespace.
 		Gid int
 		// Hostname value in UTS namespace.
 		Hostname string
+		// Register binfmt_misc entries.
+		Binfmt []BinfmtEntry
+		// Alternative pathname to attach binfmt_misc filesystem. The zero value
+		// requires [FstypeProc] to be made available at [fhs.Proc].
+		BinfmtPath *check.Absolute
 		// Sequential container setup ops.
 		*Ops
 
 		// Seccomp system call filter rules.
-		SeccompRules []seccomp.NativeRule
+		SeccompRules []std.NativeRule
 		// Extra seccomp flags.
 		SeccompFlags seccomp.ExportFlag
 		// Seccomp presets. Has no effect unless SeccompRules is zero-length.
-		SeccompPresets seccomp.FilterPreset
+		SeccompPresets std.FilterPreset
 		// Do not load seccomp program.
 		SeccompDisable bool
 
@@ -99,28 +131,114 @@ type (
 	}
 )
 
-// Start starts the container init. The init process blocks until Serve is called.
-func (p *Container) Start() error {
-	if p.cmd != nil {
-		return errors.New("container: already started")
+// A StartError contains additional information on a container startup failure.
+type StartError struct {
+	// Fatal suggests whether this error should be considered fatal for the entire program.
+	Fatal bool
+	// Step refers to the part of the setup this error is returned from.
+	Step string
+	// Err is the underlying error.
+	Err error
+	// Origin is whether this error originated from the [Container.Start] method.
+	Origin bool
+	// Passthrough is whether the Error method is passed through to Err.
+	Passthrough bool
+}
+
+func (e *StartError) Unwrap() error { return e.Err }
+func (e *StartError) Error() string {
+	if e.Passthrough {
+		return e.Err.Error()
 	}
-	if p.Ops == nil || len(*p.Ops) == 0 {
-		return errors.New("container: starting an empty container")
+	if e.Origin {
+		return e.Step
 	}
 
-	ctx, cancel := context.WithCancel(p.ctx)
-	p.cancel = cancel
+	if se, ok := errors.AsType[*os.SyscallError](e.Err); ok && se != nil {
+		return e.Step + " " + se.Error()
+	}
+
+	return e.Step + ": " + e.Err.Error()
+}
+
+// Message returns a user-facing error message.
+func (e *StartError) Message() string {
+	if e.Passthrough {
+		var (
+			numError *strconv.NumError
+		)
+
+		switch {
+		case errors.As(e.Err, new(*os.PathError)),
+			errors.As(e.Err, new(*os.SyscallError)):
+			return "cannot " + e.Err.Error()
+
+		case errors.As(e.Err, &numError) && numError != nil:
+			return "cannot parse " + strconv.Quote(numError.Num) + ": " + numError.Err.Error()
+
+		default:
+			return e.Err.Error()
+		}
+	}
+	if e.Origin {
+		return e.Step
+	}
+	return "cannot " + e.Error()
+}
+
+// for ensureCloseOnExec
+var (
+	closeOnExecOnce sync.Once
+	closeOnExecErr  error
+)
+
+// ensureCloseOnExec ensures all currently open file descriptors have the
+// syscall.FD_CLOEXEC flag set.
+//
+// This is only ran once as it is intended to handle files left open by the
+// parent, and any file opened on this side should already have
+// syscall.FD_CLOEXEC set.
+func ensureCloseOnExec() error {
+	closeOnExecOnce.Do(func() { closeOnExecErr = doCloseOnExec() })
+
+	if closeOnExecErr == nil {
+		return nil
+	}
+	return &StartError{
+		Fatal:       true,
+		Step:        "set FD_CLOEXEC on all open files",
+		Err:         closeOnExecErr,
+		Passthrough: true,
+	}
+}
+
+// Start starts the container init. The init process blocks until Serve is called.
+func (p *Container) Start() error {
+	if p == nil || p.cmd == nil ||
+		p.Ops == nil || len(*p.Ops) == 0 {
+		return errors.New("container: starting an invalid container")
+	}
+	if p.cmd.Process != nil {
+		return errors.New("container: already started")
+	}
+	if !p.InitAsRoot && len(p.Binfmt) > 0 {
+		return errors.New("container: init as root required, but not enabled")
+	}
+
+	if err := ensureCloseOnExec(); err != nil {
+		return err
+	}
 
 	// map to overflow id to work around ownership checks
 	if p.Uid < 1 {
-		p.Uid = OverflowUid()
+		p.Uid = OverflowUid(p.msg)
 	}
 	if p.Gid < 1 {
-		p.Gid = OverflowGid()
+		p.Gid = OverflowGid(p.msg)
 	}
 
 	if !p.RetainSession {
-		p.SeccompPresets |= seccomp.PresetDenyTTY
+		p.SeccompPresets |= std.PresetDenyTTY
 	}
 
 	if p.AdoptWaitDelay == 0 {
@@ -131,19 +249,26 @@ func (p *Container) Start() error {
 		p.AdoptWaitDelay = 0
 	}
 
-	p.cmd = exec.CommandContext(ctx, MustExecutable())
+	if p.cmd.Stdin == nil {
+		p.cmd.Stdin = p.Stdin
+	}
+	if p.cmd.Stdout == nil {
+		p.cmd.Stdout = p.Stdout
+	}
+	if p.cmd.Stderr == nil {
+		p.cmd.Stderr = p.Stderr
+	}
+
 	p.cmd.Args = []string{initName}
-	p.cmd.Stdin, p.cmd.Stdout, p.cmd.Stderr = p.Stdin, p.Stdout, p.Stderr
 	p.cmd.WaitDelay = p.WaitDelay
 	if p.Cancel != nil {
 		p.cmd.Cancel = func() error { return p.Cancel(p.cmd) }
 	} else {
 		p.cmd.Cancel = func() error { return p.cmd.Process.Signal(CancelSignal) }
 	}
-	p.cmd.Dir = FHSRoot
+	p.cmd.Dir = fhs.Root
 	p.cmd.SysProcAttr = &SysProcAttr{
-		Setsid:    !p.RetainSession,
-		Pdeathsig: SIGKILL,
+		Setsid: !p.RetainSession,
 		Cloneflags: CLONE_NEWUSER | CLONE_NEWPID | CLONE_NEWNS |
 			CLONE_NEWIPC | CLONE_NEWUTS | CLONE_NEWCGROUP,
 
@@ -152,25 +277,47 @@ func (p *Container) Start() error {
 			CAP_SYS_ADMIN,
 			// drop capabilities
 			CAP_SETPCAP,
+			// bring up loopback interface
+			CAP_NET_ADMIN,
 			// overlay access to upperdir and workdir
 			CAP_DAC_OVERRIDE,
 		},
 
 		UseCgroupFD: p.Cgroup != nil,
 	}
+	if !p.AllowOrphan {
+		p.cmd.SysProcAttr.Pdeathsig = SIGKILL
+	}
 	if p.cmd.SysProcAttr.UseCgroupFD {
 		p.cmd.SysProcAttr.CgroupFD = *p.Cgroup
 	}
 	if !p.HostNet {
 		p.cmd.SysProcAttr.Cloneflags |= CLONE_NEWNET
 	}
+	if p.InitAsRoot {
+		p.cmd.SysProcAttr.AmbientCaps = append(p.cmd.SysProcAttr.AmbientCaps,
+			// mappings during init as root
+			CAP_SETFCAP,
+		)
+
+		if !p.SeccompDisable &&
+			len(p.SeccompRules) == 0 &&
+			p.SeccompPresets&std.PresetDenyNS != 0 {
+			return errors.New("container: as root requires late namespace creation")
+		}
+	}
 
 	// place setup pipe before user supplied extra files, this is later restored by init
-	if fd, e, err := Setup(&p.cmd.ExtraFiles); err != nil {
-		return wrapErrSuffix(err,
-			"cannot create shim setup pipe:")
+	if r, w, err := os.Pipe(); err != nil {
+		return &StartError{
+			Fatal: true,
+			Step:  "set up params stream",
+			Err:   err,
+		}
 	} else {
-		p.setup = e
+		fd := 3 + len(p.cmd.ExtraFiles)
+		p.cmd.ExtraFiles = append(p.cmd.ExtraFiles, r)
+		p.setup[0], p.setup[1] = r, w
 		p.cmd.Env = []string{setupEnv + "=" + strconv.Itoa(fd)}
 	}
 	p.cmd.ExtraFiles = append(p.cmd.ExtraFiles, p.ExtraFiles...)
@@ -180,51 +327,63 @@ func (p *Container) Start() error {
 		runtime.LockOSThread()
 		p.wait = make(chan struct{})
 
-		done <- func() error { // setup depending on per-thread state must happen here
-			// PR_SET_NO_NEW_PRIVS: depends on per-thread state but acts on all processes created from that thread
-			if err := SetNoNewPrivs(); err != nil {
-				return wrapErrSuffix(err,
-					"prctl(PR_SET_NO_NEW_PRIVS):")
+		// setup depending on per-thread state must happen here
+		done <- func() error {
+			// PR_SET_NO_NEW_PRIVS: thread-directed but acts on all processes
+			// created from the calling thread
+			if err := setNoNewPrivs(); err != nil {
+				return &StartError{
+					Fatal: true,
+					Step:  "prctl(PR_SET_NO_NEW_PRIVS)",
+					Err:   err,
+				}
 			}
 
 			// landlock: depends on per-thread state but acts on a process group
 			{
-				rulesetAttr := &RulesetAttr{Scoped: LANDLOCK_SCOPE_SIGNAL}
+				rulesetAttr := &landlock.RulesetAttr{
+					Scoped: landlock.LANDLOCK_SCOPE_SIGNAL,
+				}
 				if !p.HostAbstract {
-					rulesetAttr.Scoped |= LANDLOCK_SCOPE_ABSTRACT_UNIX_SOCKET
+					rulesetAttr.Scoped |= landlock.LANDLOCK_SCOPE_ABSTRACT_UNIX_SOCKET
 				}
 
-				if abi, err := LandlockGetABI(); err != nil {
-					if p.HostAbstract {
-						// landlock can be skipped here as it restricts access to resources
-						// already covered by namespaces (pid)
+				if abi, err := landlock.GetABI(); err != nil {
+					if p.HostAbstract || !p.HostNet {
+						// landlock can be skipped here as it restricts access
+						// to resources already covered by namespaces (pid, net)
 						goto landlockOut
 					}
-					return wrapErrSuffix(err,
-						"landlock does not appear to be enabled:")
+					return &StartError{Step: "get landlock ABI", Err: err}
 				} else if abi < 6 {
 					if p.HostAbstract {
 						// see above comment
 						goto landlockOut
 					}
-					return msg.WrapErr(ENOSYS,
-						"kernel version too old for LANDLOCK_SCOPE_ABSTRACT_UNIX_SOCKET")
-				} else {
-					msg.Verbosef("landlock abi version %d", abi)
+					return &StartError{
+						Step:   "kernel too old for LANDLOCK_SCOPE_ABSTRACT_UNIX_SOCKET",
+						Err:    ENOSYS,
+						Origin: true,
+					}
 				}
 
 				if rulesetFd, err := rulesetAttr.Create(0); err != nil {
-					return wrapErrSuffix(err,
-						"cannot create landlock ruleset:")
+					return &StartError{
+						Fatal: true,
+						Step:  "create landlock ruleset",
+						Err:   err,
+					}
 				} else {
-					msg.Verbosef("enforcing landlock ruleset %s", rulesetAttr)
-					if err = LandlockRestrictSelf(rulesetFd, 0); err != nil {
+					if err = landlock.RestrictSelf(rulesetFd, 0); err != nil {
 						_ = Close(rulesetFd)
-						return wrapErrSuffix(err,
-							"cannot enforce landlock ruleset:")
+						return &StartError{
+							Fatal: true,
+							Step:  "enforce landlock ruleset",
+							Err:   err,
+						}
 					}
 					if err = Close(rulesetFd); err != nil {
-						msg.Verbosef("cannot close landlock ruleset: %v", err)
+						p.msg.Verbosef("cannot close landlock ruleset: %v", err)
 						// not fatal
 					}
 				}
@@ -232,9 +391,51 @@ func (p *Container) Start() error {
 			landlockOut:
 			}
 
-			msg.Verbose("starting container init")
+			// sched_setscheduler: thread-directed but acts on all processes
+			// created from the calling thread
+			if p.SetScheduler {
+				if p.SchedPolicy < 0 || p.SchedPolicy > ext.SCHED_LAST {
+					return &StartError{
+						Fatal: false,
+						Step:  "set scheduling policy",
+						Err:   EINVAL,
+					}
+				}
+
+				var param schedParam
+				if priority, err := p.SchedPolicy.GetPriorityMin(); err != nil {
+					return &StartError{
+						Fatal: true,
+						Step:  "get minimum priority",
+						Err:   err,
+					}
+				} else {
+					param.priority = max(priority, p.SchedPriority)
+				}
+
+				p.msg.Verbosef(
+					"setting scheduling policy %s priority %d",
+					p.SchedPolicy, param.priority,
+				)
+				if err := schedSetscheduler(
+					0, // calling thread
+					p.SchedPolicy,
+					&param,
+				); err != nil {
+					return &StartError{
+						Fatal: true,
+						Step:  "set scheduling policy",
+						Err:   err,
+					}
+				}
+			}
+
 			if err := p.cmd.Start(); err != nil {
-				return msg.WrapErr(err, err.Error())
+				return &StartError{
+					Step:        "start container init",
+					Err:         err,
+					Passthrough: true,
+				}
 			}
 			return nil
 		}()
@@ -246,46 +447,76 @@ func (p *Container) Start() error {
 }
 
 // Serve serves [Container.Params] to the container init.
+//
 // Serve must only be called once.
-func (p *Container) Serve() error {
-	if p.setup == nil {
+func (p *Container) Serve() (err error) {
+	if p.setup[0] == nil || p.setup[1] == nil {
 		panic("invalid serve")
 	}
 
-	setup := p.setup
-	p.setup = nil
+	done := make(chan struct{})
+	defer func() {
+		if closeErr := p.setup[1].Close(); err == nil {
+			err = closeErr
+		}
+
+		if err != nil {
+			p.cancel()
+		}
+		close(done)
+		p.setup[0], p.setup[1] = nil, nil
+	}()
+	if err = p.setup[0].Close(); err != nil {
+		return &StartError{
+			Fatal:       true,
+			Step:        "close read end of init pipe",
+			Err:         err,
+			Passthrough: true,
+		}
+	}
 
 	if p.Path == nil {
-		p.cancel()
-		return msg.WrapErr(EINVAL, "invalid executable pathname")
+		return &StartError{
+			Step:   "invalid executable pathname",
+			Err:    EINVAL,
+			Origin: true,
+		}
 	}
 
 	// do not transmit nil
 	if p.Dir == nil {
-		p.Dir = AbsFHSRoot
+		p.Dir = fhs.AbsRoot
 	}
 	if p.SeccompRules == nil {
-		p.SeccompRules = make([]seccomp.NativeRule, 0)
+		p.SeccompRules = make([]std.NativeRule, 0)
 	}
 
-	err := setup.Encode(
-		&initParams{
-			p.Params,
-			Getuid(),
-			Getgid(),
-			len(p.ExtraFiles),
-			msg.IsVerbose(),
-		},
-	)
-	if err != nil {
-		p.cancel()
-	}
-	return err
+	t := time.Now().UTC()
+	go func(f *os.File) {
+		select {
+		case <-p.ctx.Done():
+			if cancelErr := f.SetWriteDeadline(t); cancelErr != nil {
+				p.msg.Verbose(err)
+			}
+
+		case <-done:
+			return
+		}
+	}(p.setup[1])
+
+	return gob.NewEncoder(p.setup[1]).Encode(&initParams{
+		p.Params,
+		Getuid(),
+		Getgid(),
+		len(p.ExtraFiles),
+		p.msg.IsVerbose() && !p.Quiet,
+	})
 }
 
-// Wait waits for the container init process to exit and releases any resources associated with the [Container].
+// Wait blocks until the container init process to exit and releases any
+// resources associated with the [Container].
 func (p *Container) Wait() error {
-	if p.cmd == nil {
+	if p.cmd == nil || p.cmd.Process == nil {
 		return EINVAL
 	}
 
@@ -297,12 +528,44 @@ func (p *Container) Wait() error {
 	return err
 }
 
-func (p *Container) String() string {
-	return fmt.Sprintf("argv: %q, filter: %v, rules: %d, flags: %#x, presets: %#x",
-		p.Args, !p.SeccompDisable, len(p.SeccompRules), int(p.SeccompFlags), int(p.SeccompPresets))
+// StdinPipe calls the [exec.Cmd] method with the same name.
+func (p *Container) StdinPipe() (w io.WriteCloser, err error) {
+	if p.Stdin != nil {
+		return nil, errors.New("container: Stdin already set")
+	}
+	w, err = p.cmd.StdinPipe()
+	p.Stdin = p.cmd.Stdin
+	return
 }
 
-// ProcessState returns the address to os.ProcessState held by the underlying [exec.Cmd].
+// StdoutPipe calls the [exec.Cmd] method with the same name.
+func (p *Container) StdoutPipe() (r io.ReadCloser, err error) {
+	if p.Stdout != nil {
+		return nil, errors.New("container: Stdout already set")
+	}
+	r, err = p.cmd.StdoutPipe()
+	p.Stdout = p.cmd.Stdout
+	return
+}
+
+// StderrPipe calls the [exec.Cmd] method with the same name.
+func (p *Container) StderrPipe() (r io.ReadCloser, err error) {
+	if p.Stderr != nil {
+		return nil, errors.New("container: Stderr already set")
+	}
+	r, err = p.cmd.StderrPipe()
+	p.Stderr = p.cmd.Stderr
+	return
+}
+
+func (p *Container) String() string {
+	return fmt.Sprintf(
+		"argv: %q, filter: %v, rules: %d, flags: %#x, presets: %#x",
+		p.Args, !p.SeccompDisable, len(p.SeccompRules), int(p.SeccompFlags), int(p.SeccompPresets),
+	)
+}
+
+// ProcessState returns the address of os.ProcessState held by the underlying [exec.Cmd].
 func (p *Container) ProcessState() *os.ProcessState {
 	if p.cmd == nil {
 		return nil
@@ -310,14 +573,29 @@ func (p *Container) ProcessState() *os.ProcessState {
 	return p.cmd.ProcessState
 }
 
-// New returns the address to a new instance of [Container] that requires further initialisation before use.
-func New(ctx context.Context) *Container {
-	return &Container{ctx: ctx, Params: Params{Ops: new(Ops)}}
+// New returns the address to a new instance of [Container]. This value requires
+// further initialisation before use.
+func New(ctx context.Context, msg message.Msg) *Container {
+	if msg == nil {
+		msg = message.New(nil)
+	}
+
+	p := &Container{ctx: ctx, msg: msg, Params: Params{Ops: new(Ops)}}
+	c, cancel := context.WithCancel(ctx)
+	p.cancel = cancel
+	p.cmd = exec.CommandContext(c, fhs.ProcSelfExe)
+	return p
 }
 
 // NewCommand calls [New] and initialises the [Params.Path] and [Params.Args] fields.
-func NewCommand(ctx context.Context, pathname *Absolute, name string, args ...string) *Container {
-	z := New(ctx)
+func NewCommand(
+	ctx context.Context,
+	msg message.Msg,
+	pathname *check.Absolute,
+	name string,
+	args ...string,
+) *Container {
+	z := New(ctx, msg)
 	z.Path = pathname
 	z.Args = append([]string{name}, args...)
 	return z

container/container_test.go

@@ -7,25 +7,213 @@ import (
 	"errors"
 	"fmt"
 	"log"
+	"net"
 	"os"
 	"os/exec"
 	"os/signal"
+	"reflect"
 	"strconv"
 	"strings"
 	"syscall"
 	"testing"
 	"time"
+	"unsafe"
 
+	"hakurei.app/check"
 	"hakurei.app/command"
 	"hakurei.app/container"
 	"hakurei.app/container/seccomp"
-	"hakurei.app/container/vfs"
+	"hakurei.app/container/std"
+	"hakurei.app/ext"
+	"hakurei.app/fhs"
 	"hakurei.app/hst"
-	"hakurei.app/internal"
-	"hakurei.app/internal/hlog"
+	"hakurei.app/internal/info"
+	"hakurei.app/internal/landlock"
+	"hakurei.app/internal/params"
 	"hakurei.app/ldd"
+	"hakurei.app/message"
+	"hakurei.app/vfs"
 )
 
+// Note: this package requires cgo, which is unavailable in the Go playground.
+func Example() {
+	// Must be called early if the current process starts containers.
+	container.TryArgv0(nil)
+
+	// Configure the container.
+	z := container.New(context.Background(), nil)
+	z.Hostname = "hakurei-example"
+	z.Proc(fhs.AbsProc).Dev(fhs.AbsDev, true)
+	z.Stdin, z.Stdout, z.Stderr = os.Stdin, os.Stdout, os.Stderr
+
+	// Bind / for demonstration.
+	z.Bind(fhs.AbsRoot, fhs.AbsRoot, 0)
+	if name, err := exec.LookPath("hostname"); err != nil {
+		panic(err)
+	} else {
+		z.Path = check.MustAbs(name)
+	}
+
+	// This completes the first stage of container setup and starts the container init process.
+	// The new process blocks until the Serve method is called.
+	if err := z.Start(); err != nil {
+		panic(err)
+	}
+
+	// This serves the setup payload to the container init process,
+	// starting the second stage of container setup.
+	if err := z.Serve(); err != nil {
+		panic(err)
+	}
+
+	// Must be called if the Start method succeeds.
+	if err := z.Wait(); err != nil {
+		panic(err)
+	}
+
+	// Output: hakurei-example
+}
+
+func TestStartError(t *testing.T) {
+	t.Parallel()
+
+	testCases := []struct {
+		name string
+		err  error
+		s    string
+		is   error
+		isF  error
+		msg  string
+	}{
+		{"params env", &container.StartError{
+			Fatal: true,
+			Step:  "set up params stream",
+			Err:   params.ErrReceiveEnv,
+		}, "set up params stream: environment variable not set",
+			params.ErrReceiveEnv, syscall.EBADF,
+			"cannot set up params stream: environment variable not set"},
+
+		{"params", &container.StartError{
+			Fatal: true,
+			Step:  "set up params stream",
+			Err:   &os.SyscallError{Syscall: "pipe2", Err: syscall.EBADF},
+		}, "set up params stream pipe2: bad file descriptor",
+			syscall.EBADF, os.ErrInvalid,
+			"cannot set up params stream pipe2: bad file descriptor"},
+
+		{"PR_SET_NO_NEW_PRIVS", &container.StartError{
+			Fatal: true,
+			Step:  "prctl(PR_SET_NO_NEW_PRIVS)",
+			Err:   syscall.EPERM,
+		}, "prctl(PR_SET_NO_NEW_PRIVS): operation not permitted",
+			syscall.EPERM, syscall.EACCES,
+			"cannot prctl(PR_SET_NO_NEW_PRIVS): operation not permitted"},
+
+		{"landlock abi", &container.StartError{
+			Step: "get landlock ABI",
+			Err:  syscall.ENOSYS,
+		}, "get landlock ABI: function not implemented",
+			syscall.ENOSYS, syscall.ENOEXEC,
+			"cannot get landlock ABI: function not implemented"},
+
+		{"landlock old", &container.StartError{
+			Step:   "kernel version too old for LANDLOCK_SCOPE_ABSTRACT_UNIX_SOCKET",
+			Err:    syscall.ENOSYS,
+			Origin: true,
+		}, "kernel version too old for LANDLOCK_SCOPE_ABSTRACT_UNIX_SOCKET",
+			syscall.ENOSYS, syscall.ENOSPC,
+			"kernel version too old for LANDLOCK_SCOPE_ABSTRACT_UNIX_SOCKET"},
+
+		{"landlock create", &container.StartError{
+			Fatal: true,
+			Step:  "create landlock ruleset",
+			Err:   syscall.EBADFD,
+		}, "create landlock ruleset: file descriptor in bad state",
+			syscall.EBADFD, syscall.EBADF,
+			"cannot create landlock ruleset: file descriptor in bad state"},
+
+		{"landlock enforce", &container.StartError{
+			Fatal: true,
+			Step:  "enforce landlock ruleset",
+			Err:   syscall.ENOTRECOVERABLE,
+		}, "enforce landlock ruleset: state not recoverable",
+			syscall.ENOTRECOVERABLE, syscall.ETIMEDOUT,
+			"cannot enforce landlock ruleset: state not recoverable"},
+
+		{"start", &container.StartError{
+			Step: "start container init",
+			Err: &os.PathError{
+				Op:   "fork/exec",
+				Path: "/proc/nonexistent",
+				Err:  syscall.ENOENT,
+			}, Passthrough: true,
+		}, "fork/exec /proc/nonexistent: no such file or directory",
+			syscall.ENOENT, syscall.ENOSYS,
+			"cannot fork/exec /proc/nonexistent: no such file or directory"},
+
+		{"start syscall", &container.StartError{
+			Step: "start container init",
+			Err: &os.SyscallError{
+				Syscall: "open",
+				Err:     syscall.ENOSYS,
+			}, Passthrough: true,
+		}, "open: function not implemented",
+			syscall.ENOSYS, syscall.ENOENT,
+			"cannot open: function not implemented"},
+
+		{"start FD_CLOEXEC", &container.StartError{
+			Fatal:       true,
+			Step:        "set FD_CLOEXEC on all open files",
+			Err:         func() error { _, err := strconv.Atoi("invalid"); return err }(),
+			Passthrough: true,
+		}, `strconv.Atoi: parsing "invalid": invalid syntax`,
+			strconv.ErrSyntax, os.ErrInvalid,
+			`cannot parse "invalid": invalid syntax`},
+
+		{"start other", &container.StartError{
+			Step: "start container init",
+			Err: &net.OpError{
+				Op:  "dial",
+				Net: "unix",
+				Err: syscall.ECONNREFUSED,
+			}, Passthrough: true,
+		}, "dial unix: connection refused",
+			syscall.ECONNREFUSED, syscall.ECONNABORTED,
+			"dial unix: connection refused"},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			t.Run("error", func(t *testing.T) {
+				if got := tc.err.Error(); got != tc.s {
+					t.Errorf("Error: %q, want %q", got, tc.s)
+				}
+			})
+
+			t.Run("is", func(t *testing.T) {
+				if !errors.Is(tc.err, tc.is) {
+					t.Error("Is: unexpected false")
+				}
+				if errors.Is(tc.err, tc.isF) {
+					t.Errorf("Is: unexpected true")
+				}
+			})
+
+			t.Run("msg", func(t *testing.T) {
+				if got, ok := message.GetMessage(tc.err); !ok {
+					if tc.msg != "" {
+						t.Errorf("GetMessage: err does not implement MessageError")
+					}
+					return
+				} else if got != tc.msg {
+					t.Errorf("GetMessage: %q, want %q", got, tc.msg)
+				}
+			})
+		})
+	}
+}
+
 const (
 	ignore  = "\x00"
 	ignoreV = -1
@@ -47,6 +235,9 @@ func earlyMnt(mnt ...*vfs.MountInfoEntry) func(*testing.T, context.Context) []*v
 	return func(*testing.T, context.Context) []*vfs.MountInfoEntry { return mnt }
 }
 
+//go:linkname toHost hakurei.app/container.toHost
+func toHost(name string) string
+
 var containerTestCases = []struct {
 	name    string
 	filter  bool
@@ -60,83 +251,83 @@ var containerTestCases = []struct {
 	uid int
 	gid int
 
-	rules   []seccomp.NativeRule
+	rules   []std.NativeRule
 	flags   seccomp.ExportFlag
-	presets seccomp.FilterPreset
+	presets std.FilterPreset
 }{
 	{"minimal", true, false, false, true,
 		emptyOps, emptyMnt,
-		1000, 100, nil, 0, seccomp.PresetStrict},
+		1000, 100, nil, 0, std.PresetStrict},
 	{"allow", true, true, true, false,
 		emptyOps, emptyMnt,
-		1000, 100, nil, 0, seccomp.PresetExt | seccomp.PresetDenyDevel},
+		1000, 100, nil, 0, std.PresetExt | std.PresetDenyDevel},
 	{"no filter", false, true, true, true,
 		emptyOps, emptyMnt,
-		1000, 100, nil, 0, seccomp.PresetExt},
+		1000, 100, nil, 0, std.PresetExt},
 	{"custom rules", true, true, true, false,
 		emptyOps, emptyMnt,
-		1, 31, []seccomp.NativeRule{{seccomp.ScmpSyscall(syscall.SYS_SETUID), seccomp.ScmpErrno(syscall.EPERM), nil}}, 0, seccomp.PresetExt},
+		1, 31, []std.NativeRule{{Syscall: ext.SyscallNum(syscall.SYS_SETUID), Errno: std.ScmpErrno(syscall.EPERM)}}, 0, std.PresetExt},
 
 	{"tmpfs", true, false, false, true,
 		earlyOps(new(container.Ops).
-			Tmpfs(hst.AbsTmp, 0, 0755),
+			Tmpfs(hst.AbsPrivateTmp, 0, 0755),
 		),
 		earlyMnt(
-			ent("/", hst.Tmp, "rw,nosuid,nodev,relatime", "tmpfs", "ephemeral", ignore),
+			ent("/", hst.PrivateTmp, "rw,nosuid,nodev,relatime", "tmpfs", "ephemeral", ignore),
 		),
-		9, 9, nil, 0, seccomp.PresetStrict},
+		9, 9, nil, 0, std.PresetStrict},
 
 	{"dev", true, true /* go test output is not a tty */, false, false,
 		earlyOps(new(container.Ops).
-			Dev(container.MustAbs("/dev"), true),
+			Dev(check.MustAbs("/dev"), true),
 		),
 		earlyMnt(
-			ent("/", "/dev", "ro,nosuid,nodev,relatime", "tmpfs", "devtmpfs", ignore),
-			ent("/null", "/dev/null", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
-			ent("/zero", "/dev/zero", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
-			ent("/full", "/dev/full", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
-			ent("/random", "/dev/random", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
-			ent("/urandom", "/dev/urandom", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
-			ent("/tty", "/dev/tty", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
+			ent("/", "/dev", "ro,nosuid,nodev,relatime", "tmpfs", ignore, ignore),
+			ent("/null", "/dev/null", ignore, "devtmpfs", ignore, ignore),
+			ent("/zero", "/dev/zero", ignore, "devtmpfs", ignore, ignore),
+			ent("/full", "/dev/full", ignore, "devtmpfs", ignore, ignore),
+			ent("/random", "/dev/random", ignore, "devtmpfs", ignore, ignore),
+			ent("/urandom", "/dev/urandom", ignore, "devtmpfs", ignore, ignore),
+			ent("/tty", "/dev/tty", ignore, "devtmpfs", ignore, ignore),
 			ent("/", "/dev/pts", "rw,nosuid,noexec,relatime", "devpts", "devpts", "rw,mode=620,ptmxmode=666"),
 			ent("/", "/dev/mqueue", "rw,nosuid,nodev,noexec,relatime", "mqueue", "mqueue", "rw"),
 			ent("/", "/dev/shm", "rw,nosuid,nodev,relatime", "tmpfs", "tmpfs", ignore),
 		),
-		1971, 100, nil, 0, seccomp.PresetStrict},
+		1971, 100, nil, 0, std.PresetStrict},
 
 	{"dev no mqueue", true, true /* go test output is not a tty */, false, false,
 		earlyOps(new(container.Ops).
-			Dev(container.MustAbs("/dev"), false),
+			Dev(check.MustAbs("/dev"), false),
 		),
 		earlyMnt(
-			ent("/", "/dev", "ro,nosuid,nodev,relatime", "tmpfs", "devtmpfs", ignore),
-			ent("/null", "/dev/null", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
-			ent("/zero", "/dev/zero", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
-			ent("/full", "/dev/full", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
-			ent("/random", "/dev/random", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
-			ent("/urandom", "/dev/urandom", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
-			ent("/tty", "/dev/tty", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
+			ent("/", "/dev", "ro,nosuid,nodev,relatime", "tmpfs", ignore, ignore),
+			ent("/null", "/dev/null", ignore, "devtmpfs", ignore, ignore),
+			ent("/zero", "/dev/zero", ignore, "devtmpfs", ignore, ignore),
+			ent("/full", "/dev/full", ignore, "devtmpfs", ignore, ignore),
+			ent("/random", "/dev/random", ignore, "devtmpfs", ignore, ignore),
+			ent("/urandom", "/dev/urandom", ignore, "devtmpfs", ignore, ignore),
+			ent("/tty", "/dev/tty", ignore, "devtmpfs", ignore, ignore),
 			ent("/", "/dev/pts", "rw,nosuid,noexec,relatime", "devpts", "devpts", "rw,mode=620,ptmxmode=666"),
 			ent("/", "/dev/shm", "rw,nosuid,nodev,relatime", "tmpfs", "tmpfs", ignore),
 		),
-		1971, 100, nil, 0, seccomp.PresetStrict},
+		1971, 100, nil, 0, std.PresetStrict},
 
 	{"overlay", true, false, false, true,
 		func(t *testing.T) (*container.Ops, context.Context) {
-			tempDir := container.MustAbs(t.TempDir())
+			tempDir := check.MustAbs(t.TempDir())
 			lower0, lower1, upper, work :=
 				tempDir.Append("lower0"),
 				tempDir.Append("lower1"),
 				tempDir.Append("upper"),
 				tempDir.Append("work")
-			for _, a := range []*container.Absolute{lower0, lower1, upper, work} {
+			for _, a := range []*check.Absolute{lower0, lower1, upper, work} {
 				if err := os.Mkdir(a.String(), 0755); err != nil {
 					t.Fatalf("Mkdir: error = %v", err)
 				}
 			}
 
 			return new(container.Ops).
-					Overlay(hst.AbsTmp, upper, work, lower0, lower1),
+					Overlay(hst.AbsPrivateTmp, upper, work, lower0, lower1),
 				context.WithValue(context.WithValue(context.WithValue(context.WithValue(t.Context(),
 					testVal("lower1"), lower1),
 					testVal("lower0"), lower0),
@@ -145,112 +336,91 @@ var containerTestCases = []struct {
 		},
 		func(t *testing.T, ctx context.Context) []*vfs.MountInfoEntry {
 			return []*vfs.MountInfoEntry{
-				ent("/", hst.Tmp, "rw", "overlay", "overlay",
-					"rw,lowerdir="+
-						container.InternalToHostOvlEscape(ctx.Value(testVal("lower0")).(*container.Absolute).String())+":"+
-						container.InternalToHostOvlEscape(ctx.Value(testVal("lower1")).(*container.Absolute).String())+
+				ent("/", hst.PrivateTmp, "rw", "overlay", "overlay",
+					"rw"+
+						",lowerdir+="+
+						toHost(ctx.Value(testVal("lower0")).(*check.Absolute).String())+
+						",lowerdir+="+
+						toHost(ctx.Value(testVal("lower1")).(*check.Absolute).String())+
 						",upperdir="+
-						container.InternalToHostOvlEscape(ctx.Value(testVal("upper")).(*container.Absolute).String())+
+						toHost(ctx.Value(testVal("upper")).(*check.Absolute).String())+
 						",workdir="+
-						container.InternalToHostOvlEscape(ctx.Value(testVal("work")).(*container.Absolute).String())+
+						toHost(ctx.Value(testVal("work")).(*check.Absolute).String())+
 						",redirect_dir=nofollow,uuid=on,userxattr"),
 			}
 		},
-		1 << 3, 1 << 14, nil, 0, seccomp.PresetStrict},
+		1 << 3, 1 << 14, nil, 0, std.PresetStrict},
 
 	{"overlay ephemeral", true, false, false, true,
 		func(t *testing.T) (*container.Ops, context.Context) {
-			tempDir := container.MustAbs(t.TempDir())
+			tempDir := check.MustAbs(t.TempDir())
 			lower0, lower1 :=
 				tempDir.Append("lower0"),
 				tempDir.Append("lower1")
-			for _, a := range []*container.Absolute{lower0, lower1} {
+			for _, a := range []*check.Absolute{lower0, lower1} {
 				if err := os.Mkdir(a.String(), 0755); err != nil {
 					t.Fatalf("Mkdir: error = %v", err)
 				}
 			}
 
 			return new(container.Ops).
-					OverlayEphemeral(hst.AbsTmp, lower0, lower1),
+					OverlayEphemeral(hst.AbsPrivateTmp, lower0, lower1),
 				t.Context()
 		},
 		func(t *testing.T, ctx context.Context) []*vfs.MountInfoEntry {
 			return []*vfs.MountInfoEntry{
 				// contains random suffix
-				ent("/", hst.Tmp, "rw", "overlay", "overlay", ignore),
+				ent("/", hst.PrivateTmp, "rw", "overlay", "overlay", ignore),
 			}
 		},
-		1 << 3, 1 << 14, nil, 0, seccomp.PresetStrict},
+		1 << 3, 1 << 14, nil, 0, std.PresetStrict},
 
 	{"overlay readonly", true, false, false, true,
 		func(t *testing.T) (*container.Ops, context.Context) {
-			tempDir := container.MustAbs(t.TempDir())
+			tempDir := check.MustAbs(t.TempDir())
 			lower0, lower1 :=
 				tempDir.Append("lower0"),
 				tempDir.Append("lower1")
-			for _, a := range []*container.Absolute{lower0, lower1} {
+			for _, a := range []*check.Absolute{lower0, lower1} {
 				if err := os.Mkdir(a.String(), 0755); err != nil {
 					t.Fatalf("Mkdir: error = %v", err)
 				}
 			}
 			return new(container.Ops).
-					OverlayReadonly(hst.AbsTmp, lower0, lower1),
+					OverlayReadonly(hst.AbsPrivateTmp, lower0, lower1),
 				context.WithValue(context.WithValue(t.Context(),
 					testVal("lower1"), lower1),
 					testVal("lower0"), lower0)
 		},
 		func(t *testing.T, ctx context.Context) []*vfs.MountInfoEntry {
 			return []*vfs.MountInfoEntry{
-				ent("/", hst.Tmp, "rw", "overlay", "overlay",
-					"ro,lowerdir="+
-						container.InternalToHostOvlEscape(ctx.Value(testVal("lower0")).(*container.Absolute).String())+":"+
-						container.InternalToHostOvlEscape(ctx.Value(testVal("lower1")).(*container.Absolute).String())+
+				ent("/", hst.PrivateTmp, "rw", "overlay", "overlay",
+					"ro"+
+						",lowerdir+="+
+						toHost(ctx.Value(testVal("lower0")).(*check.Absolute).String())+
+						",lowerdir+="+
+						toHost(ctx.Value(testVal("lower1")).(*check.Absolute).String())+
 						",redirect_dir=nofollow,userxattr"),
 			}
 		},
-		1 << 3, 1 << 14, nil, 0, seccomp.PresetStrict},
+		1 << 3, 1 << 14, nil, 0, std.PresetStrict},
 }
 
 func TestContainer(t *testing.T) {
-	replaceOutput(t)
-
-	t.Run("cancel", testContainerCancel(nil, func(t *testing.T, c *container.Container) {
-		wantErr := context.Canceled
-		wantExitCode := 0
-		if err := c.Wait(); !errors.Is(err, wantErr) {
-			container.GetOutput().PrintBaseErr(err, "wait:")
-			t.Errorf("Wait: error = %v, want %v", err, wantErr)
-		}
-		if ps := c.ProcessState(); ps == nil {
-			t.Errorf("ProcessState unexpectedly returned nil")
-		} else if code := ps.ExitCode(); code != wantExitCode {
-			t.Errorf("ExitCode: %d, want %d", code, wantExitCode)
-		}
-	}))
-
-	t.Run("forward", testContainerCancel(func(c *container.Container) {
-		c.ForwardCancel = true
-	}, func(t *testing.T, c *container.Container) {
-		var exitError *exec.ExitError
-		if err := c.Wait(); !errors.As(err, &exitError) {
-			container.GetOutput().PrintBaseErr(err, "wait:")
-			t.Errorf("Wait: error = %v", err)
-		}
-		if code := exitError.ExitCode(); code != blockExitCodeInterrupt {
-			t.Errorf("ExitCode: %d, want %d", code, blockExitCodeInterrupt)
-		}
-	}))
+	t.Parallel()
 
+	var suffix string
+runTests:
 	for i, tc := range containerTestCases {
-		t.Run(tc.name, func(t *testing.T) {
+		_suffix := suffix
+		t.Run(tc.name+_suffix, func(t *testing.T) {
+			t.Parallel()
+
 			wantOps, wantOpsCtx := tc.ops(t)
 			wantMnt := tc.mnt(t, wantOpsCtx)
 
-			ctx, cancel := context.WithTimeout(t.Context(), helperDefaultTimeout)
-			defer cancel()
-
-			var libPaths []*container.Absolute
-			c := helperNewContainerLibPaths(ctx, &libPaths, "container", strconv.Itoa(i))
+			var libPaths []*check.Absolute
+			c := helperNewContainerLibPaths(t.Context(), &libPaths, "container", strconv.Itoa(i))
 			c.Uid = tc.uid
 			c.Gid = tc.gid
 			c.Hostname = hostnameFromTestCase(tc.name)
@@ -260,7 +430,6 @@ func TestContainer(t *testing.T) {
 			} else {
 				c.Stdout, c.Stderr = os.Stdout, os.Stderr
 			}
-			c.WaitDelay = helperDefaultTimeout
 			*c.Ops = append(*c.Ops, *wantOps...)
 			c.SeccompRules = tc.rules
 			c.SeccompFlags = tc.flags | seccomp.AllowMultiarch
@@ -268,13 +437,27 @@ func TestContainer(t *testing.T) {
 			c.SeccompDisable = !tc.filter
 			c.RetainSession = tc.session
 			c.HostNet = tc.net
+			c.InitAsRoot = _suffix != ""
+			c.Env = append(c.Env, "HAKUREI_TEST_SUFFIX="+_suffix)
+			if info.CanDegrade {
+				if _, err := landlock.GetABI(); err != nil {
+					if !errors.Is(err, syscall.ENOSYS) {
+						t.Fatalf("LandlockGetABI: error = %v", err)
+					}
+					c.HostAbstract = true
+					t.Log("Landlock LSM is unavailable, enabling HostAbstract")
+				}
+			}
+			if c.InitAsRoot {
+				c.SeccompPresets &= ^std.PresetDenyNS
+			}
 
 			c.
-				Readonly(container.MustAbs(pathReadonly), 0755).
-				Tmpfs(container.MustAbs("/tmp"), 0, 0755).
-				Place(container.MustAbs("/etc/hostname"), []byte(c.Hostname))
+				Readonly(check.MustAbs(pathReadonly), 0755).
+				Tmpfs(check.MustAbs("/tmp"), 0, 0755).
+				Place(check.MustAbs("/etc/hostname"), []byte(c.Hostname))
 			// needs /proc to check mountinfo
-			c.Proc(container.MustAbs("/proc"))
+			c.Proc(check.MustAbs("/proc"))
 
 			// mountinfo cannot be resolved directly by helper due to libPaths nondeterminism
 			mnt := make([]*vfs.MountInfoEntry, 0, 3+len(libPaths))
@@ -305,28 +488,42 @@ func TestContainer(t *testing.T) {
 				_, _ = output.WriteTo(os.Stdout)
 				t.Fatalf("cannot serialise expected mount points: %v", err)
 			}
-			c.Place(container.MustAbs(pathWantMnt), want.Bytes())
+			c.Place(check.MustAbs(pathWantMnt), want.Bytes())
 
 			if tc.ro {
-				c.Remount(container.MustAbs("/"), syscall.MS_RDONLY)
+				c.Remount(check.MustAbs("/"), syscall.MS_RDONLY)
 			}
 
 			if err := c.Start(); err != nil {
 				_, _ = output.WriteTo(os.Stdout)
-				container.GetOutput().PrintBaseErr(err, "start:")
-				t.Fatalf("cannot start container: %v", err)
+				if m, ok := container.InternalMessageFromError(err); ok {
+					t.Fatal(m)
+				} else {
+					t.Fatalf("cannot start container: %v", err)
+				}
 			} else if err = c.Serve(); err != nil {
 				_, _ = output.WriteTo(os.Stdout)
-				container.GetOutput().PrintBaseErr(err, "serve:")
-				t.Errorf("cannot serve setup params: %v", err)
+				if m, ok := container.InternalMessageFromError(err); ok {
+					t.Error(m)
+				} else {
+					t.Errorf("cannot serve setup params: %v", err)
+				}
 			}
 			if err := c.Wait(); err != nil {
 				_, _ = output.WriteTo(os.Stdout)
-				container.GetOutput().PrintBaseErr(err, "wait:")
-				t.Fatalf("wait: %v", err)
+				if m, ok := container.InternalMessageFromError(err); ok {
+					t.Fatal(m)
+				} else {
+					t.Fatalf("wait: %v", err)
+				}
 			}
 		})
 	}
+
+	if suffix == "" {
+		suffix = " as root"
+		goto runTests
+	}
 }
 
 func ent(root, target, vfsOptstr, fsType, source, fsOptstr string) *vfs.MountInfoEntry {
@@ -349,52 +546,129 @@ func hostnameFromTestCase(name string) string {
 }
 
 func testContainerCancel(
+	t *testing.T,
 	containerExtra func(c *container.Container),
-	waitCheck func(t *testing.T, c *container.Container),
-) func(t *testing.T) {
-	return func(t *testing.T) {
-		ctx, cancel := context.WithTimeout(t.Context(), helperDefaultTimeout)
+	waitCheck func(ps *os.ProcessState, waitErr error),
+) {
+	ctx, cancel := context.WithCancel(t.Context())
 
-		c := helperNewContainer(ctx, "block")
-		c.Stdout, c.Stderr = os.Stdout, os.Stderr
-		c.WaitDelay = helperDefaultTimeout
-		if containerExtra != nil {
-			containerExtra(c)
+	c := helperNewContainer(ctx, "block")
+	c.Stdout, c.Stderr = os.Stdout, os.Stderr
+	if containerExtra != nil {
+		containerExtra(c)
+	}
+
+	ready := make(chan struct{})
+	var waitErr error
+	r, w, err := os.Pipe()
+	if err != nil {
+		t.Fatalf("cannot pipe: %v", err)
+	}
+
+	c.ExtraFiles = append(c.ExtraFiles, w)
+	go func() {
+		defer close(ready)
+		if _, _err := r.Read(make([]byte, 1)); _err != nil {
+			panic(_err)
 		}
+	}()
 
-		ready := make(chan struct{})
-		if r, w, err := os.Pipe(); err != nil {
-			t.Fatalf("cannot pipe: %v", err)
+	if err = c.Start(); err != nil {
+		if m, ok := container.InternalMessageFromError(err); ok {
+			t.Fatal(m)
 		} else {
-			c.ExtraFiles = append(c.ExtraFiles, w)
-			go func() {
-				defer close(ready)
-				if _, err = r.Read(make([]byte, 1)); err != nil {
-					panic(err.Error())
-				}
-			}()
-		}
-
-		if err := c.Start(); err != nil {
-			container.GetOutput().PrintBaseErr(err, "start:")
 			t.Fatalf("cannot start container: %v", err)
-		} else if err = c.Serve(); err != nil {
-			container.GetOutput().PrintBaseErr(err, "serve:")
+		}
+	}
+
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		waitErr = c.Wait()
+		_ = r.SetReadDeadline(time.Now())
+	}()
+
+	if err = c.Serve(); err != nil {
+		if m, ok := container.InternalMessageFromError(err); ok {
+			t.Error(m)
+		} else {
 			t.Errorf("cannot serve setup params: %v", err)
 		}
-		<-ready
-		cancel()
-		waitCheck(t, c)
 	}
+	<-ready
+	cancel()
+	<-done
+	waitCheck(c.ProcessState(), waitErr)
+}
+
+func TestForward(t *testing.T) {
+	t.Parallel()
+
+	f := func(ps *os.ProcessState, waitErr error) {
+		var exitError *exec.ExitError
+		if !errors.As(waitErr, &exitError) {
+			if m, ok := container.InternalMessageFromError(waitErr); ok {
+				t.Error(m)
+			}
+			t.Errorf("Wait: error = %v", waitErr)
+		}
+		if code := exitError.ExitCode(); code != blockExitCodeInterrupt {
+			t.Errorf("ExitCode: %d, want %d", code, blockExitCodeInterrupt)
+		}
+	}
+	t.Run("direct", func(t *testing.T) {
+		t.Parallel()
+		testContainerCancel(t, func(c *container.Container) {
+			c.ForwardCancel = true
+		}, f)
+	})
+	t.Run("as root", func(t *testing.T) {
+		testContainerCancel(t, func(c *container.Container) {
+			c.ForwardCancel = true
+			c.InitAsRoot = true
+			c.Proc(fhs.AbsProc)
+		}, f)
+	})
+}
+
+func TestCancel(t *testing.T) {
+	t.Parallel()
+
+	f := func(ps *os.ProcessState, waitErr error) {
+		wantErr := context.Canceled
+		if !reflect.DeepEqual(waitErr, wantErr) {
+			if m, ok := container.InternalMessageFromError(waitErr); ok {
+				t.Error(m)
+			}
+			t.Errorf("Wait: error = %#v, want %#v", waitErr, wantErr)
+		}
+		if ps == nil {
+			t.Errorf("ProcessState unexpectedly returned nil")
+		} else if code := ps.ExitCode(); code != 0 {
+			t.Errorf("ExitCode: %d, want %d", code, 0)
+		}
+	}
+	t.Run("direct", func(t *testing.T) {
+		t.Parallel()
+		testContainerCancel(t, nil, f)
+	})
+	t.Run("as root", func(t *testing.T) {
+		testContainerCancel(t, func(c *container.Container) {
+			c.InitAsRoot = true
+			c.Proc(fhs.AbsProc)
+		}, f)
+	})
 }
 
 func TestContainerString(t *testing.T) {
-	c := container.NewCommand(t.Context(), container.MustAbs("/run/current-system/sw/bin/ldd"), "ldd", "/usr/bin/env")
+	t.Parallel()
+	msg := message.New(nil)
+	c := container.NewCommand(t.Context(), msg, check.MustAbs("/run/current-system/sw/bin/ldd"), "ldd", "/usr/bin/env")
 	c.SeccompFlags |= seccomp.AllowMultiarch
 	c.SeccompRules = seccomp.Preset(
-		seccomp.PresetExt|seccomp.PresetDenyNS|seccomp.PresetDenyTTY,
+		std.PresetExt|std.PresetDenyNS|std.PresetDenyTTY,
 		c.SeccompFlags)
-	c.SeccompPresets = seccomp.PresetStrict
+	c.SeccompPresets = std.PresetStrict
 	want := `argv: ["ldd" "/usr/bin/env"], filter: true, rules: 65, flags: 0x1, presets: 0xf`
 	if got := c.String(); got != want {
 		t.Errorf("String: %s, want %s", got, want)
@@ -408,18 +682,19 @@ const (
 func init() {
 	helperCommands = append(helperCommands, func(c command.Command) {
 		c.Command("block", command.UsageInternal, func(args []string) error {
+			sig := make(chan os.Signal, 1)
+			signal.Notify(sig, os.Interrupt)
+			go func() { <-sig; os.Exit(blockExitCodeInterrupt) }()
+
 			if _, err := os.NewFile(3, "sync").Write([]byte{0}); err != nil {
 				return fmt.Errorf("write to sync pipe: %v", err)
 			}
-			{
-				sig := make(chan os.Signal, 1)
-				signal.Notify(sig, os.Interrupt)
-				go func() { <-sig; os.Exit(blockExitCodeInterrupt) }()
-			}
 			select {}
 		})
 
 		c.Command("container", command.UsageInternal, func(args []string) error {
+			asRoot := os.Getenv("HAKUREI_TEST_SUFFIX") == " as root"
+
 			if len(args) != 1 {
 				return syscall.EINVAL
 			}
@@ -437,6 +712,66 @@ func init() {
 				return fmt.Errorf("gid: %d, want %d", gid, tc.gid)
 			}
 
+			// no attack surface increase during as root due to no_new_privs
+			var wantBounding uintptr = 1
+			asRootNot := " not"
+			if !asRoot {
+				wantBounding = 0
+				asRootNot = ""
+			}
+
+			const (
+				PR_CAP_AMBIENT        = 0x2f
+				PR_CAP_AMBIENT_IS_SET = 0x1
+			)
+			for i := range container.LastCap(nil) + 1 {
+				r, _, errno := syscall.Syscall(
+					syscall.SYS_PRCTL,
+					PR_CAP_AMBIENT,
+					PR_CAP_AMBIENT_IS_SET,
+					i,
+				)
+				if errno != 0 {
+					return os.NewSyscallError("prctl", errno)
+				}
+				if r != 0 {
+					return fmt.Errorf("capability %d in ambient set", i)
+				}
+
+				r, _, errno = syscall.Syscall(
+					syscall.SYS_PRCTL,
+					syscall.PR_CAPBSET_READ,
+					i,
+					0,
+				)
+				if errno != 0 {
+					return os.NewSyscallError("prctl", errno)
+				}
+				if r != wantBounding {
+					return fmt.Errorf("capability %d%s in bounding set", i, asRootNot)
+				}
+			}
+
+			const _LINUX_CAPABILITY_VERSION_3 = 0x20080522
+			var capData struct {
+				effective   uint32
+				permitted   uint32
+				inheritable uint32
+			}
+			if _, _, errno := syscall.Syscall(syscall.SYS_CAPGET, uintptr(unsafe.Pointer(&struct {
+				version uint32
+				pid     int32
+			}{_LINUX_CAPABILITY_VERSION_3, 0})), uintptr(unsafe.Pointer(&capData)), 0); errno != 0 {
+				return os.NewSyscallError("capget", errno)
+			}
+
+			if max(capData.effective, capData.permitted, capData.inheritable) != 0 {
+				return fmt.Errorf(
+					"effective = %d, permitted = %d, inheritable = %d",
+					capData.effective, capData.permitted, capData.inheritable,
+				)
+			}
+
 			wantHost := hostnameFromTestCase(tc.name)
 			if host, err := os.Hostname(); err != nil {
 				return fmt.Errorf("cannot get hostname: %v", err)
@@ -484,11 +819,22 @@ func init() {
 						return fmt.Errorf("got more than %d entries", len(mnt))
 					}
 
-					// ugly hack but should be reliable and is less likely to false negative than comparing by parsed flags
-					cur.VfsOptstr = strings.TrimSuffix(cur.VfsOptstr, ",relatime")
-					cur.VfsOptstr = strings.TrimSuffix(cur.VfsOptstr, ",noatime")
-					mnt[i].VfsOptstr = strings.TrimSuffix(mnt[i].VfsOptstr, ",relatime")
-					mnt[i].VfsOptstr = strings.TrimSuffix(mnt[i].VfsOptstr, ",noatime")
+					// ugly hack but should be reliable and is less likely to
+					//false negative than comparing by parsed flags
+					for _, s := range []string{
+						"relatime",
+						"noatime",
+					} {
+						cur.VfsOptstr = strings.TrimSuffix(cur.VfsOptstr, ","+s)
+						mnt[i].VfsOptstr = strings.TrimSuffix(mnt[i].VfsOptstr, ","+s)
+					}
+					for _, s := range []string{
+						"seclabel",
+						"inode64",
+					} {
+						cur.FsOptstr = strings.Replace(cur.FsOptstr, ","+s, "", 1)
+						mnt[i].FsOptstr = strings.Replace(mnt[i].FsOptstr, ","+s, "", 1)
+					}
 
 					if !cur.EqualWithIgnore(mnt[i], "\x00") {
 						fail = true
@@ -520,18 +866,17 @@ func init() {
 const (
 	envDoCheck = "HAKUREI_TEST_DO_CHECK"
 
-	helperDefaultTimeout = 5 * time.Second
-	helperInnerPath      = "/usr/bin/helper"
+	helperInnerPath = "/usr/bin/helper"
 )
 
 var (
-	absHelperInnerPath = container.MustAbs(helperInnerPath)
+	absHelperInnerPath = check.MustAbs(helperInnerPath)
 )
 
 var helperCommands []func(c command.Command)
 
 func TestMain(m *testing.M) {
-	container.TryArgv0(hlog.Output{}, hlog.Prepare, internal.InstallOutput)
+	container.TryArgv0(nil)
 
 	if os.Getenv(envDoCheck) == "1" {
 		c := command.New(os.Stderr, log.Printf, "helper", func(args []string) error {
@@ -544,7 +889,7 @@ func TestMain(m *testing.M) {
 		}
 		c.MustParse(os.Args[1:], func(err error) {
 			if err != nil {
-				log.Fatal(err.Error())
+				log.Fatal(err)
 			}
 		})
 		return
@@ -553,13 +898,16 @@ func TestMain(m *testing.M) {
 	os.Exit(m.Run())
 }
 
-func helperNewContainerLibPaths(ctx context.Context, libPaths *[]*container.Absolute, args ...string) (c *container.Container) {
-	c = container.NewCommand(ctx, absHelperInnerPath, "helper", args...)
+func helperNewContainerLibPaths(ctx context.Context, libPaths *[]*check.Absolute, args ...string) (c *container.Container) {
+	msg := message.New(nil)
+	msg.SwapVerbose(testing.Verbose())
+
+	c = container.NewCommand(ctx, msg, absHelperInnerPath, "helper", args...)
 	c.Env = append(c.Env, envDoCheck+"=1")
-	c.Bind(container.MustAbs(os.Args[0]), absHelperInnerPath, 0)
+	c.Bind(fhs.AbsProcSelfExe, absHelperInnerPath, 0)
 
 	// in case test has cgo enabled
-	if entries, err := ldd.Exec(ctx, os.Args[0]); err != nil {
+	if entries, err := ldd.Resolve(ctx, msg, nil); err != nil {
 		log.Fatalf("ldd: %v", err)
 	} else {
 		*libPaths = ldd.Path(entries)
@@ -572,5 +920,5 @@ func helperNewContainerLibPaths(ctx context.Context, libPaths *[]*container.Abso
 }
 
 func helperNewContainer(ctx context.Context, args ...string) (c *container.Container) {
-	return helperNewContainerLibPaths(ctx, new([]*container.Absolute), args...)
+	return helperNewContainerLibPaths(ctx, new([]*check.Absolute), args...)
 }

container/dispatcher.go

@@ -1,9 +1,10 @@
 package container
 
 import (
+	"context"
 	"io"
 	"io/fs"
-	"log"
+	"net"
 	"os"
 	"os/exec"
 	"os/signal"
@@ -12,6 +13,11 @@ import (
 	"syscall"
 
 	"hakurei.app/container/seccomp"
+	"hakurei.app/container/std"
+	"hakurei.app/ext"
+	"hakurei.app/internal/netlink"
+	"hakurei.app/internal/params"
+	"hakurei.app/message"
 )
 
 type osFile interface {
@@ -20,7 +26,8 @@ type osFile interface {
 	fs.File
 }
 
-// syscallDispatcher provides methods that make state-dependent system calls as part of their behaviour.
+// syscallDispatcher provides methods that make state-dependent system calls as
+// part of their behaviour.
 type syscallDispatcher interface {
 	// new starts a goroutine with a new instance of syscallDispatcher.
 	// A syscallDispatcher must never be used in any goroutine other than the one owning it,
@@ -38,7 +45,7 @@ type syscallDispatcher interface {
 	setNoNewPrivs() error
 
 	// lastcap provides [LastCap].
-	lastcap() uintptr
+	lastcap(msg message.Msg) uintptr
 	// capset provides capset.
 	capset(hdrp *capHeader, datap *[2]capData) error
 	// capBoundingSetDrop provides capBoundingSetDrop.
@@ -50,19 +57,23 @@ type syscallDispatcher interface {
 	// isatty provides [Isatty].
 	isatty(fd int) bool
 	// receive provides [Receive].
-	receive(key string, e any, fdp *uintptr) (closeFunc func() error, err error)
+	receive(key string, e any, fdp *int) (closeFunc func() error, err error)
 
 	// bindMount provides procPaths.bindMount.
-	bindMount(source, target string, flags uintptr, eq bool) error
+	bindMount(msg message.Msg, source, target string, flags uintptr) error
 	// remount provides procPaths.remount.
-	remount(target string, flags uintptr) error
+	remount(msg message.Msg, target string, flags uintptr) error
 	// mountTmpfs provides mountTmpfs.
 	mountTmpfs(fsname, target string, flags uintptr, size int, perm os.FileMode) error
+	// mountOverlay provides mountOverlay.
+	mountOverlay(target string, options [][2]string) error
 	// ensureFile provides ensureFile.
 	ensureFile(name string, perm, pperm os.FileMode) error
+	// mustLoopback provides mustLoopback.
+	mustLoopback(ctx context.Context, msg message.Msg)
 
 	// seccompLoad provides [seccomp.Load].
-	seccompLoad(rules []seccomp.NativeRule, flags seccomp.ExportFlag) error
+	seccompLoad(rules []std.NativeRule, flags seccomp.ExportFlag) error
 	// notify provides [signal.Notify].
 	notify(c chan<- os.Signal, sig ...os.Signal)
 	// start starts [os/exec.Cmd].
@@ -122,24 +133,12 @@ type syscallDispatcher interface {
 	// wait4 provides syscall.Wait4
 	wait4(pid int, wstatus *syscall.WaitStatus, options int, rusage *syscall.Rusage) (wpid int, err error)
 
-	// printf provides [log.Printf].
-	printf(format string, v ...any)
-	// fatal provides [log.Fatal]
-	fatal(v ...any)
-	// fatalf provides [log.Fatalf]
-	fatalf(format string, v ...any)
-	// verbose provides [Msg.Verbose].
-	verbose(v ...any)
-	// verbosef provides [Msg.Verbosef].
-	verbosef(format string, v ...any)
-	// suspend provides [Msg.Suspend].
-	suspend()
-	// resume provides [Msg.Resume].
-	resume() bool
-	// beforeExit provides [Msg.BeforeExit].
-	beforeExit()
-	// printBaseErr provides [Msg.PrintBaseErr].
-	printBaseErr(err error, fallback string)
+	// printf provides the Printf method of [log.Logger].
+	printf(msg message.Msg, format string, v ...any)
+	// fatal provides the Fatal method of [log.Logger]
+	fatal(msg message.Msg, v ...any)
+	// fatalf provides the Fatalf method of [log.Logger]
+	fatalf(msg message.Msg, format string, v ...any)
 }
 
 // direct implements syscallDispatcher on the current kernel.
@@ -149,34 +148,81 @@ func (k direct) new(f func(k syscallDispatcher)) { go f(k) }
 
 func (direct) lockOSThread() { runtime.LockOSThread() }
 
-func (direct) setPtracer(pid uintptr) error       { return SetPtracer(pid) }
-func (direct) setDumpable(dumpable uintptr) error { return SetDumpable(dumpable) }
-func (direct) setNoNewPrivs() error               { return SetNoNewPrivs() }
+func (direct) setPtracer(pid uintptr) error       { return ext.SetPtracer(pid) }
+func (direct) setDumpable(dumpable uintptr) error { return ext.SetDumpable(dumpable) }
+func (direct) setNoNewPrivs() error               { return setNoNewPrivs() }
 
-func (direct) lastcap() uintptr                                { return LastCap() }
+func (direct) lastcap(msg message.Msg) uintptr                 { return LastCap(msg) }
 func (direct) capset(hdrp *capHeader, datap *[2]capData) error { return capset(hdrp, datap) }
 func (direct) capBoundingSetDrop(cap uintptr) error            { return capBoundingSetDrop(cap) }
 func (direct) capAmbientClearAll() error                       { return capAmbientClearAll() }
 func (direct) capAmbientRaise(cap uintptr) error               { return capAmbientRaise(cap) }
-func (direct) isatty(fd int) bool                              { return Isatty(fd) }
-func (direct) receive(key string, e any, fdp *uintptr) (func() error, error) {
-	return Receive(key, e, fdp)
+func (direct) isatty(fd int) bool                              { return ext.Isatty(fd) }
+func (direct) receive(key string, e any, fdp *int) (func() error, error) {
+	return params.Receive(key, e, fdp)
 }
 
-func (direct) bindMount(source, target string, flags uintptr, eq bool) error {
-	return hostProc.bindMount(source, target, flags, eq)
+func (direct) bindMount(msg message.Msg, source, target string, flags uintptr) error {
+	return hostProc.bindMount(msg, source, target, flags)
 }
-func (direct) remount(target string, flags uintptr) error {
-	return hostProc.remount(target, flags)
+func (direct) remount(msg message.Msg, target string, flags uintptr) error {
+	return hostProc.remount(msg, target, flags)
 }
 func (k direct) mountTmpfs(fsname, target string, flags uintptr, size int, perm os.FileMode) error {
 	return mountTmpfs(k, fsname, target, flags, size, perm)
 }
+func (k direct) mountOverlay(target string, options [][2]string) error {
+	return mountOverlay(target, options)
+}
 func (direct) ensureFile(name string, perm, pperm os.FileMode) error {
 	return ensureFile(name, perm, pperm)
 }
+func (direct) mustLoopback(ctx context.Context, msg message.Msg) {
+	var lo int
+	if ifi, err := net.InterfaceByName("lo"); err != nil {
+		msg.GetLogger().Fatalln(err)
+	} else {
+		lo = ifi.Index
+	}
 
-func (direct) seccompLoad(rules []seccomp.NativeRule, flags seccomp.ExportFlag) error {
+	c, err := netlink.DialRoute(0)
+	if err != nil {
+		msg.GetLogger().Fatalln(err)
+	}
+
+	must := func(err error) {
+		if err == nil {
+			return
+		}
+		if closeErr := c.Close(); closeErr != nil {
+			msg.Verbosef("cannot close RTNETLINK: %v", closeErr)
+		}
+
+		switch err.(type) {
+		case *os.SyscallError:
+			msg.GetLogger().Fatalf("cannot %v", err)
+
+		case syscall.Errno:
+			msg.GetLogger().Fatalf("RTNETLINK answers: %v", err)
+
+		default:
+			if err == context.DeadlineExceeded || err == context.Canceled {
+				msg.GetLogger().Fatalf("interrupted RTNETLINK operation")
+			}
+			msg.GetLogger().Fatal("RTNETLINK answers with malformed message")
+		}
+	}
+	must(c.SendNewaddrLo(ctx, uint32(lo)))
+	must(c.SendIfInfomsg(ctx, syscall.RTM_NEWLINK, 0, &syscall.IfInfomsg{
+		Family: syscall.AF_UNSPEC,
+		Index:  int32(lo),
+		Flags:  syscall.IFF_UP,
+		Change: syscall.IFF_UP,
+	}))
+	must(c.Close())
+}
+
+func (direct) seccompLoad(rules []std.NativeRule, flags seccomp.ExportFlag) error {
 	return seccomp.Load(rules, flags)
 }
 func (direct) notify(c chan<- os.Signal, sig ...os.Signal) { signal.Notify(c, sig...) }
@@ -225,7 +271,7 @@ func (direct) pivotRoot(newroot, putold string) (err error) {
 	return syscall.PivotRoot(newroot, putold)
 }
 func (direct) mount(source, target, fstype string, flags uintptr, data string) (err error) {
-	return syscall.Mount(source, target, fstype, flags, data)
+	return mount(source, target, fstype, flags, data)
 }
 func (direct) unmount(target string, flags int) (err error) {
 	return syscall.Unmount(target, flags)
@@ -234,12 +280,6 @@ func (direct) wait4(pid int, wstatus *syscall.WaitStatus, options int, rusage *s
 	return syscall.Wait4(pid, wstatus, options, rusage)
 }
 
-func (direct) printf(format string, v ...any)          { log.Printf(format, v...) }
-func (direct) fatal(v ...any)                          { log.Fatal(v...) }
-func (direct) fatalf(format string, v ...any)          { log.Fatalf(format, v...) }
-func (direct) verbose(v ...any)                        { msg.Verbose(v...) }
-func (direct) verbosef(format string, v ...any)        { msg.Verbosef(format, v...) }
-func (direct) suspend()                                { msg.Suspend() }
-func (direct) resume() bool                            { return msg.Resume() }
-func (direct) beforeExit()                             { msg.BeforeExit() }
-func (direct) printBaseErr(err error, fallback string) { msg.PrintBaseErr(err, fallback) }
+func (direct) printf(msg message.Msg, format string, v ...any) { msg.GetLogger().Printf(format, v...) }
+func (direct) fatal(msg message.Msg, v ...any)                 { msg.GetLogger().Fatal(v...) }
+func (direct) fatalf(msg message.Msg, format string, v ...any) { msg.GetLogger().Fatalf(format, v...) }