hst: expose scheduling priority

This is useful when limits are configured to allow it.

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

cmd/hakurei/command.go

@@ -16,6 +16,7 @@ import (
 	"hakurei.app/command"
 	"hakurei.app/container/check"
 	"hakurei.app/container/fhs"
+	"hakurei.app/container/std"
 	"hakurei.app/hst"
 	"hakurei.app/internal/dbus"
 	"hakurei.app/internal/env"
@@ -89,6 +90,9 @@ func buildCommand(ctx context.Context, msg message.Msg, early *earlyHardeningErr
 			flagHomeDir  string
 			flagUserName string
 
+			flagSchedPolicy   string
+			flagSchedPriority int
+
 			flagPrivateRuntime, flagPrivateTmpdir bool
 
 			flagWayland, flagX11, flagDBus, flagPipeWire, flagPulse bool
@@ -131,7 +135,7 @@ func buildCommand(ctx context.Context, msg message.Msg, early *earlyHardeningErr
 					log.Fatal(optionalErrorUnwrap(err))
 					return err
 				} else if progPath, err = check.NewAbs(p); err != nil {
-					log.Fatal(err.Error())
+					log.Fatal(err)
 					return err
 				}
 			}
@@ -150,7 +154,7 @@ func buildCommand(ctx context.Context, msg message.Msg, early *earlyHardeningErr
 				et |= hst.EPipeWire
 			}
 
-			config := &hst.Config{
+			config := hst.Config{
 				ID:          flagID,
 				Identity:    flagIdentity,
 				Groups:      flagGroups,
@@ -177,6 +181,13 @@ func buildCommand(ctx context.Context, msg message.Msg, early *earlyHardeningErr
 				},
 			}
 
+			if err := config.SchedPolicy.UnmarshalText(
+				[]byte(flagSchedPolicy),
+			); err != nil {
+				log.Fatal(err)
+			}
+			config.SchedPriority = std.Int(flagSchedPriority)
+
 			// bind GPU stuff
 			if et&(hst.EX11|hst.EWayland) != 0 {
 				config.Container.Filesystem = append(config.Container.Filesystem, hst.FilesystemConfigJSON{FilesystemConfig: &hst.FSBind{
@@ -214,7 +225,7 @@ func buildCommand(ctx context.Context, msg message.Msg, early *earlyHardeningErr
 					homeDir = passwd.HomeDir
 				}
 				if a, err := check.NewAbs(homeDir); err != nil {
-					log.Fatal(err.Error())
+					log.Fatal(err)
 					return err
 				} else {
 					config.Container.Home = a
@@ -234,11 +245,11 @@ func buildCommand(ctx context.Context, msg message.Msg, early *earlyHardeningErr
 					config.SessionBus = dbus.NewConfig(flagID, true, flagDBusMpris)
 				} else {
 					if f, err := os.Open(flagDBusConfigSession); err != nil {
-						log.Fatal(err.Error())
+						log.Fatal(err)
 					} else {
 						decodeJSON(log.Fatal, "load session bus proxy config", f, &config.SessionBus)
 						if err = f.Close(); err != nil {
-							log.Fatal(err.Error())
+							log.Fatal(err)
 						}
 					}
 				}
@@ -246,11 +257,11 @@ func buildCommand(ctx context.Context, msg message.Msg, early *earlyHardeningErr
 				// system bus proxy is optional
 				if flagDBusConfigSystem != "nil" {
 					if f, err := os.Open(flagDBusConfigSystem); err != nil {
-						log.Fatal(err.Error())
+						log.Fatal(err)
 					} else {
 						decodeJSON(log.Fatal, "load system bus proxy config", f, &config.SystemBus)
 						if err = f.Close(); err != nil {
-							log.Fatal(err.Error())
+							log.Fatal(err)
 						}
 					}
 				}
@@ -266,7 +277,7 @@ func buildCommand(ctx context.Context, msg message.Msg, early *earlyHardeningErr
 				}
 			}
 
-			outcome.Main(ctx, msg, config, -1)
+			outcome.Main(ctx, msg, &config, -1)
 			panic("unreachable")
 		}).
 			Flag(&flagDBusConfigSession, "dbus-config", command.StringFlag("builtin"),
@@ -287,6 +298,10 @@ func buildCommand(ctx context.Context, msg message.Msg, early *earlyHardeningErr
 				"Container home directory").
 			Flag(&flagUserName, "u", command.StringFlag("chronos"),
 				"Passwd user name within sandbox").
+			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),

cmd/hakurei/command_test.go

@@ -36,7 +36,7 @@ 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>] [--private-runtime] [--private-tmpdir] [--wayland] [-X] [--dbus] [--pipewire] [--pulse] COMMAND [OPTIONS]
+Usage:	hakurei run [-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
@@ -60,6 +60,10 @@ Flags:
     	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

cmd/mbf/main.go

@@ -87,7 +87,7 @@ func main() {
 		}
 
 		if flagIdle {
-			pkg.SchedPolicy = container.SCHED_IDLE
+			pkg.SetSchedIdle = true
 		}
 
 		return

container/container.go

@@ -38,9 +38,13 @@ type (
 	Container struct {
 		// Whether the container init should stay alive after its parent terminates.
 		AllowOrphan bool
-		// Scheduling policy to set via sched_setscheduler(2). The zero value
-		// skips this call. Supported policies are [SCHED_BATCH], [SCHED_IDLE].
-		SchedPolicy int
+		// Whether to set SchedPolicy and SchedPriority via sched_setscheduler(2).
+		SetScheduler bool
+		// Scheduling policy to set via sched_setscheduler(2).
+		SchedPolicy std.SchedPolicy
+		// Scheduling priority to set via sched_setscheduler(2). The zero value
+		// implies the minimum value supported by the current SchedPolicy.
+		SchedPriority std.Int
 		// Cgroup fd, nil to disable.
 		Cgroup *int
 		// ExtraFiles passed through to initial process in the container, with
@@ -373,16 +377,38 @@ func (p *Container) Start() error {
 
 			// sched_setscheduler: thread-directed but acts on all processes
 			// created from the calling thread
-			if p.SchedPolicy > 0 {
-				p.msg.Verbosef("setting scheduling policy %d", p.SchedPolicy)
+			if p.SetScheduler {
+				if p.SchedPolicy < 0 || p.SchedPolicy > std.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,
-					&schedParam{0},
+					&param,
 				); err != nil {
 					return &StartError{
 						Fatal: true,
-						Step:  "enforce landlock ruleset",
+						Step:  "set scheduling policy",
 						Err:   err,
 					}
 				}

container/std/syscall.go

@@ -1,6 +1,12 @@
 package std
 
-import "iter"
+import (
+	"encoding"
+	"iter"
+	"strconv"
+	"sync"
+	"syscall"
+)
 
 // Syscalls returns an iterator over all wired syscalls.
 func Syscalls() iter.Seq2[string, ScmpSyscall] {
@@ -26,3 +32,128 @@ func SyscallResolveName(name string) (num ScmpSyscall, ok bool) {
 	num, ok = syscallNumExtra[name]
 	return
 }
+
+// SchedPolicy denotes a scheduling policy defined in include/uapi/linux/sched.h.
+type SchedPolicy int
+
+// include/uapi/linux/sched.h
+const (
+	SCHED_NORMAL SchedPolicy = iota
+	SCHED_FIFO
+	SCHED_RR
+	SCHED_BATCH
+	_SCHED_ISO // SCHED_ISO: reserved but not implemented yet
+	SCHED_IDLE
+	SCHED_DEADLINE
+	SCHED_EXT
+
+	SCHED_LAST SchedPolicy = iota - 1
+)
+
+var _ encoding.TextMarshaler = SCHED_LAST
+var _ encoding.TextUnmarshaler = new(SCHED_LAST)
+
+// String returns a unique representation of policy, also used in encoding.
+func (policy SchedPolicy) String() string {
+	switch policy {
+	case SCHED_NORMAL:
+		return ""
+	case SCHED_FIFO:
+		return "fifo"
+	case SCHED_RR:
+		return "rr"
+	case SCHED_BATCH:
+		return "batch"
+	case SCHED_IDLE:
+		return "idle"
+	case SCHED_DEADLINE:
+		return "deadline"
+	case SCHED_EXT:
+		return "ext"
+
+	default:
+		return "invalid policy " + strconv.Itoa(int(policy))
+	}
+}
+
+// MarshalText performs bounds checking and returns the result of String.
+func (policy SchedPolicy) MarshalText() ([]byte, error) {
+	if policy == _SCHED_ISO || policy < 0 || policy > SCHED_LAST {
+		return nil, syscall.EINVAL
+	}
+	return []byte(policy.String()), nil
+}
+
+// InvalidSchedPolicyError is an invalid string representation of a [SchedPolicy].
+type InvalidSchedPolicyError string
+
+func (InvalidSchedPolicyError) Unwrap() error { return syscall.EINVAL }
+func (e InvalidSchedPolicyError) Error() string {
+	return "invalid scheduling policy " + strconv.Quote(string(e))
+}
+
+// UnmarshalText is the inverse of MarshalText.
+func (policy *SchedPolicy) UnmarshalText(text []byte) error {
+	switch string(text) {
+	case "fifo":
+		*policy = SCHED_FIFO
+	case "rr":
+		*policy = SCHED_RR
+	case "batch":
+		*policy = SCHED_BATCH
+	case "idle":
+		*policy = SCHED_IDLE
+	case "deadline":
+		*policy = SCHED_DEADLINE
+	case "ext":
+		*policy = SCHED_EXT
+
+	case "":
+		*policy = 0
+		return nil
+	default:
+		return InvalidSchedPolicyError(text)
+	}
+	return nil
+}
+
+// for sched_get_priority_max and sched_get_priority_min
+var (
+	schedPriority     [SCHED_LAST + 1][2]Int
+	schedPriorityErr  [SCHED_LAST + 1][2]error
+	schedPriorityOnce [SCHED_LAST + 1][2]sync.Once
+)
+
+// GetPriorityMax returns the maximum priority value that can be used with the
+// scheduling algorithm identified by policy.
+func (policy SchedPolicy) GetPriorityMax() (Int, error) {
+	schedPriorityOnce[policy][0].Do(func() {
+		priority, _, errno := syscall.Syscall(
+			syscall.SYS_SCHED_GET_PRIORITY_MAX,
+			uintptr(policy),
+			0, 0,
+		)
+		schedPriority[policy][0] = Int(priority)
+		if errno != 0 {
+			schedPriorityErr[policy][0] = errno
+		}
+	})
+	return schedPriority[policy][0], schedPriorityErr[policy][0]
+}
+
+// GetPriorityMin returns the minimum priority value that can be used with the
+// scheduling algorithm identified by policy.
+func (policy SchedPolicy) GetPriorityMin() (Int, error) {
+	schedPriorityOnce[policy][1].Do(func() {
+		priority, _, errno := syscall.Syscall(
+			syscall.SYS_SCHED_GET_PRIORITY_MIN,
+			uintptr(policy),
+			0, 0,
+		)
+		schedPriority[policy][1] = Int(priority)
+		if errno != 0 {
+			schedPriorityErr[policy][1] = errno
+		}
+	})
+	return schedPriority[policy][1], schedPriorityErr[policy][1]
+}

container/std/syscall_test.go

@@ -1,6 +1,11 @@
 package std_test
 
 import (
+	"encoding/json"
+	"errors"
+	"math"
+	"reflect"
+	"syscall"
 	"testing"
 
 	"hakurei.app/container/std"
@@ -19,3 +24,90 @@ func TestSyscallResolveName(t *testing.T) {
 		})
 	}
 }
+
+func TestSchedPolicyJSON(t *testing.T) {
+	t.Parallel()
+
+	testCases := []struct {
+		policy    std.SchedPolicy
+		want      string
+		encodeErr error
+		decodeErr error
+	}{
+		{std.SCHED_NORMAL, `""`, nil, nil},
+		{std.SCHED_FIFO, `"fifo"`, nil, nil},
+		{std.SCHED_RR, `"rr"`, nil, nil},
+		{std.SCHED_BATCH, `"batch"`, nil, nil},
+		{4, `"invalid policy 4"`, syscall.EINVAL, std.InvalidSchedPolicyError("invalid policy 4")},
+		{std.SCHED_IDLE, `"idle"`, nil, nil},
+		{std.SCHED_DEADLINE, `"deadline"`, nil, nil},
+		{std.SCHED_EXT, `"ext"`, nil, nil},
+		{math.MaxInt, `"iso"`, syscall.EINVAL, std.InvalidSchedPolicyError("iso")},
+	}
+	for _, tc := range testCases {
+		name := tc.policy.String()
+		if tc.policy == std.SCHED_NORMAL {
+			name = "normal"
+		}
+
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+
+			got, err := json.Marshal(tc.policy)
+			if !errors.Is(err, tc.encodeErr) {
+				t.Fatalf("Marshal: error = %v, want %v", err, tc.encodeErr)
+			}
+			if err == nil && string(got) != tc.want {
+				t.Fatalf("Marshal: %s, want %s", string(got), tc.want)
+			}
+
+			var v std.SchedPolicy
+			if err = json.Unmarshal([]byte(tc.want), &v); !reflect.DeepEqual(err, tc.decodeErr) {
+				t.Fatalf("Unmarshal: error = %v, want %v", err, tc.decodeErr)
+			}
+			if err == nil && v != tc.policy {
+				t.Fatalf("Unmarshal: %d, want %d", v, tc.policy)
+			}
+		})
+	}
+}
+
+func TestSchedPolicyMinMax(t *testing.T) {
+	t.Parallel()
+
+	testCases := []struct {
+		policy   std.SchedPolicy
+		min, max std.Int
+		err      error
+	}{
+		{std.SCHED_NORMAL, 0, 0, nil},
+		{std.SCHED_FIFO, 1, 99, nil},
+		{std.SCHED_RR, 1, 99, nil},
+		{std.SCHED_BATCH, 0, 0, nil},
+		{4, -1, -1, syscall.EINVAL},
+		{std.SCHED_IDLE, 0, 0, nil},
+		{std.SCHED_DEADLINE, 0, 0, nil},
+		{std.SCHED_EXT, 0, 0, nil},
+	}
+	for _, tc := range testCases {
+		name := tc.policy.String()
+		if tc.policy == std.SCHED_NORMAL {
+			name = "normal"
+		}
+
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+
+			if priority, err := tc.policy.GetPriorityMax(); !reflect.DeepEqual(err, tc.err) {
+				t.Fatalf("GetPriorityMax: error = %v, want %v", err, tc.err)
+			} else if priority != tc.max {
+				t.Fatalf("GetPriorityMax: %d, want %d", priority, tc.max)
+			}
+			if priority, err := tc.policy.GetPriorityMin(); !reflect.DeepEqual(err, tc.err) {
+				t.Fatalf("GetPriorityMin: error = %v, want %v", err, tc.err)
+			} else if priority != tc.min {
+				t.Fatalf("GetPriorityMin: %d, want %d", priority, tc.min)
+			}
+		})
+	}
+}

container/syscall.go

@@ -43,18 +43,6 @@ func Isatty(fd int) bool {
 	return r == 0
 }
 
-// include/uapi/linux/sched.h
-const (
-	SCHED_NORMAL = iota
-	SCHED_FIFO
-	SCHED_RR
-	SCHED_BATCH
-	_ // SCHED_ISO: reserved but not implemented yet
-	SCHED_IDLE
-	SCHED_DEADLINE
-	SCHED_EXT
-)
-
 // schedParam is equivalent to struct sched_param from include/linux/sched.h.
 type schedParam struct {
 	// sched_priority
@@ -74,13 +62,13 @@ type schedParam struct {
 // this if you do not have something similar in place!
 //
 // [very subtle to use correctly]: https://www.openwall.com/lists/musl/2016/03/01/4
-func schedSetscheduler(tid, policy int, param *schedParam) error {
-	if r, _, errno := Syscall(
+func schedSetscheduler(tid int, policy std.SchedPolicy, param *schedParam) error {
+	if _, _, errno := Syscall(
 		SYS_SCHED_SETSCHEDULER,
 		uintptr(tid),
 		uintptr(policy),
 		uintptr(unsafe.Pointer(param)),
-	); r < 0 {
+	); errno != 0 {
 		return errno
 	}
 	return nil

flake.lock

@@ -7,11 +7,11 @@
         ]
       },
       "locked": {
-        "lastModified": 1765384171,
-        "narHash": "sha256-FuFtkJrW1Z7u+3lhzPRau69E0CNjADku1mLQQflUORo=",
+        "lastModified": 1772985280,
+        "narHash": "sha256-FdrNykOoY9VStevU4zjSUdvsL9SzJTcXt4omdEDZDLk=",
         "owner": "nix-community",
         "repo": "home-manager",
-        "rev": "44777152652bc9eacf8876976fa72cc77ca8b9d8",
+        "rev": "8f736f007139d7f70752657dff6a401a585d6cbc",
         "type": "github"
       },
       "original": {
@@ -23,11 +23,11 @@
     },
     "nixpkgs": {
       "locked": {
-        "lastModified": 1765311797,
-        "narHash": "sha256-mSD5Ob7a+T2RNjvPvOA1dkJHGVrNVl8ZOrAwBjKBDQo=",
+        "lastModified": 1772822230,
+        "narHash": "sha256-yf3iYLGbGVlIthlQIk5/4/EQDZNNEmuqKZkQssMljuw=",
         "owner": "NixOS",
         "repo": "nixpkgs",
-        "rev": "09eb77e94fa25202af8f3e81ddc7353d9970ac1b",
+        "rev": "71caefce12ba78d84fe618cf61644dce01cf3a96",
         "type": "github"
       },
       "original": {

flake.nix

@@ -99,7 +99,7 @@
           hakurei = pkgs.pkgsStatic.callPackage ./package.nix {
             inherit (pkgs)
               # passthru.buildInputs
-              go
+              go_1_26
               clang
 
               # nativeBuildInputs
@@ -182,7 +182,7 @@
                         let
                           # this is used for interactive vm testing during development, where tests might be broken
                           package = self.packages.${pkgs.stdenv.hostPlatform.system}.hakurei.override {
-                            buildGoModule = previousArgs: pkgs.pkgsStatic.buildGoModule (previousArgs // { doCheck = false; });
+                            buildGo126Module = previousArgs: pkgs.pkgsStatic.buildGo126Module (previousArgs // { doCheck = false; });
                           };
                         in
                         {

go.mod

@@ -1,3 +1,3 @@
 module hakurei.app
 
-go 1.25
+go 1.26

hst/config.go

@@ -6,96 +6,137 @@ import (
 	"strings"
 
 	"hakurei.app/container/check"
+	"hakurei.app/container/std"
 )
 
-// Config configures an application container, implemented in internal/app.
+// Config configures an application container.
 type Config struct {
 	// Reverse-DNS style configured arbitrary identifier string.
-	// Passed to wayland security-context-v1 and used as part of defaults in dbus session proxy.
+	//
+	// This value is passed as is to Wayland security-context-v1 and used as
+	// part of defaults in D-Bus session proxy. The zero value causes a default
+	// value to be derived from the container instance.
 	ID string `json:"id,omitempty"`
 
 	// System services to make available in the container.
 	Enablements *Enablements `json:"enablements,omitempty"`
 
 	// Session D-Bus proxy configuration.
-	// If set to nil, session bus proxy assume built-in defaults.
+	//
+	// Has no effect if [EDBus] but is not set in Enablements. The zero value
+	// assumes built-in defaults derived from ID.
 	SessionBus *BusConfig `json:"session_bus,omitempty"`
 	// System D-Bus proxy configuration.
-	// If set to nil, system bus proxy is disabled.
+	//
+	// Has no effect if [EDBus] but is not set in Enablements. The zero value
+	// disables system bus proxy.
 	SystemBus *BusConfig `json:"system_bus,omitempty"`
 
-	// Direct access to wayland socket, no attempt is made to attach security-context-v1
-	// and the bare socket is made available to the container.
+	// Direct access to Wayland socket, no attempt is made to attach
+	// security-context-v1 and the bare socket is made available to the
+	// container.
 	//
-	// This option is unsupported and most likely enables full control over the Wayland
-	// session. Do not set this to true unless you are sure you know what you are doing.
+	// This option is unsupported and will most likely enable full control over
+	// the Wayland session from within the container. Do not set this to true
+	// unless you are sure you know what you are doing.
 	DirectWayland bool `json:"direct_wayland,omitempty"`
-	// Direct access to the PipeWire socket established via SecurityContext::Create, no
-	// attempt is made to start the pipewire-pulse server.
+
+	// Direct access to the PipeWire socket established via SecurityContext::Create,
+	// no attempt is made to start the pipewire-pulse server.
 	//
-	// The SecurityContext machinery is fatally flawed, it blindly sets read and execute
-	// bits on all objects for clients with the lowest achievable privilege level (by
-	// setting PW_KEY_ACCESS to "restricted"). This enables them to call any method
-	// targeting any object, and since Registry::Destroy checks for the read and execute bit,
-	// allows the destruction of any object other than PW_ID_CORE as well. This behaviour
-	// is implemented separately in media-session and wireplumber, with the wireplumber
-	// implementation in Lua via an embedded Lua vm. In all known setups, wireplumber is
-	// in use, and there is no known way to change its behaviour and set permissions
-	// differently without replacing the Lua script. Also, since PipeWire relies on these
-	// permissions to work, reducing them is not possible.
+	// The SecurityContext machinery is fatally flawed, it unconditionally sets
+	// read and execute bits on all objects for clients with the lowest achievable
+	// privilege level (by setting PW_KEY_ACCESS to "restricted" or by satisfying
+	// all conditions of [the /.flatpak-info hack]). This enables them to call
+	// any method targeting any object, and since Registry::Destroy checks for
+	// the read and execute bit, allows the destruction of any object other than
+	// PW_ID_CORE as well.
 	//
-	// Currently, the only other sandboxed use case is flatpak, which is not aware of
-	// PipeWire and blindly exposes the bare PulseAudio socket to the container (behaves
-	// like DirectPulse). This socket is backed by the pipewire-pulse compatibility daemon,
-	// which obtains client pid via the SO_PEERCRED option. The PipeWire daemon, pipewire-pulse
-	// daemon and the session manager daemon then separately performs the /.flatpak-info hack
-	// described in https://git.gensokyo.uk/security/hakurei/issues/21. Under such use case,
-	// since the client has no direct access to PipeWire, insecure parts of the protocol are
-	// obscured by pipewire-pulse simply not implementing them, and thus hiding the flaws
-	// described above.
+	// This behaviour is implemented separately in media-session and wireplumber,
+	// with the wireplumber implementation in Lua via an embedded Lua vm. In all
+	// known setups, wireplumber is in use, and in that case, no option for
+	// configuring this behaviour exists, without replacing the Lua script.
+	// Also, since PipeWire relies on these permissions to work, reducing them
+	// was never possible in the first place.
 	//
-	// Hakurei does not rely on the /.flatpak-info hack. Instead, a socket is sets up via
-	// SecurityContext. A pipewire-pulse server connected through it achieves the same
-	// permissions as flatpak does via the /.flatpak-info hack and is maintained for the
-	// life of the container.
+	// Currently, the only other sandboxed use case is flatpak, which is not
+	// aware of PipeWire and blindly exposes the bare PulseAudio socket to the
+	// container (behaves like DirectPulse). This socket is backed by the
+	// pipewire-pulse compatibility daemon, which obtains client pid via the
+	// SO_PEERCRED option. The PipeWire daemon, pipewire-pulse daemon and the
+	// session manager daemon then separately performs [the /.flatpak-info hack].
+	// Under such use case, since the client has no direct access to PipeWire,
+	// insecure parts of the protocol are obscured by the absence of an
+	// equivalent API in PulseAudio, or pipewire-pulse simply not implementing
+	// them.
+	//
+	// Hakurei does not rely on [the /.flatpak-info hack]. Instead, a socket is
+	// sets up via SecurityContext. A pipewire-pulse server connected through it
+	// achieves the same permissions as flatpak does via [the /.flatpak-info hack]
+	// and is maintained for the life of the container.
+	//
+	// This option is unsupported and enables a denial-of-service attack as the
+	// sandboxed client is able to destroy any client object and thus
+	// disconnecting them from PipeWire, or destroy the SecurityContext object,
+	// preventing any further container creation.
 	//
-	// This option is unsupported and enables a denial-of-service attack as the sandboxed
-	// client is able to destroy any client object and thus disconnecting them from PipeWire,
-	// or destroy the SecurityContext object preventing any further container creation.
 	// Do not set this to true, it is insecure under any configuration.
-	DirectPipeWire bool `json:"direct_pipewire,omitempty"`
-	// Direct access to PulseAudio socket, no attempt is made to establish pipewire-pulse
-	// server via a PipeWire socket with a SecurityContext attached and the bare socket
-	// is made available to the container.
 	//
-	// This option is unsupported and enables arbitrary code execution as the PulseAudio
-	// server. Do not set this to true, it is insecure under any configuration.
+	// [the /.flatpak-info hack]: https://git.gensokyo.uk/security/hakurei/issues/21
+	DirectPipeWire bool `json:"direct_pipewire,omitempty"`
+
+	// Direct access to PulseAudio socket, no attempt is made to establish
+	// pipewire-pulse server via a PipeWire socket with a SecurityContext
+	// attached, and the bare socket is made available to the container.
+	//
+	// This option is unsupported and enables arbitrary code execution as the
+	// PulseAudio server.
+	//
+	// Do not set this to true, it is insecure under any configuration.
 	DirectPulse bool `json:"direct_pulse,omitempty"`
 
 	// Extra acl updates to perform before setuid.
 	ExtraPerms []ExtraPermConfig `json:"extra_perms,omitempty"`
 
-	// Numerical application id, passed to hsu, used to derive init user namespace credentials.
+	// Numerical application id, passed to hsu, used to derive init user
+	// namespace credentials.
 	Identity int `json:"identity"`
 	// Init user namespace supplementary groups inherited by all container processes.
 	Groups []string `json:"groups"`
 
+	// Scheduling policy to set for the container.
+	//
+	// The zero value retains the current scheduling policy.
+	SchedPolicy std.SchedPolicy `json:"sched_policy,omitempty"`
+	// Scheduling priority to set for the container.
+	//
+	// The zero value implies the minimum priority of the current SchedPolicy.
+	// Has no effect if SchedPolicy is zero.
+	SchedPriority std.Int `json:"sched_priority,omitempty"`
+
 	// High level configuration applied to the underlying [container].
 	Container *ContainerConfig `json:"container"`
 }
 
 var (
-	// ErrConfigNull is returned by [Config.Validate] for an invalid configuration that contains a null value for any
-	// field that must not be null.
+	// ErrConfigNull is returned by [Config.Validate] for an invalid configuration
+	// that contains a null value for any field that must not be null.
 	ErrConfigNull = errors.New("unexpected null in config")
 
-	// ErrIdentityBounds is returned by [Config.Validate] for an out of bounds [Config.Identity] value.
+	// ErrIdentityBounds is returned by [Config.Validate] for an out of bounds
+	// [Config.Identity] value.
 	ErrIdentityBounds = errors.New("identity out of bounds")
 
-	// ErrEnviron is returned by [Config.Validate] if an environment variable name contains '=' or NUL.
+	// ErrSchedPolicyBounds is returned by [Config.Validate] for an out of bounds
+	// [Config.SchedPolicy] value.
+	ErrSchedPolicyBounds = errors.New("scheduling policy out of bounds")
+
+	// ErrEnviron is returned by [Config.Validate] if an environment variable
+	// name contains '=' or NUL.
 	ErrEnviron = errors.New("invalid environment variable name")
 
-	// ErrInsecure is returned by [Config.Validate] if the configuration is considered insecure.
+	// ErrInsecure is returned by [Config.Validate] if the configuration is
+	// considered insecure.
 	ErrInsecure = errors.New("configuration is insecure")
 )
 
@@ -112,6 +153,13 @@ func (config *Config) Validate() error {
 			Msg: "identity " + strconv.Itoa(config.Identity) + " out of range"}
 	}
 
+	if config.SchedPolicy < 0 || config.SchedPolicy > std.SCHED_LAST {
+		return &AppError{Step: "validate configuration", Err: ErrSchedPolicyBounds,
+			Msg: "scheduling policy " +
+				strconv.Itoa(int(config.SchedPolicy)) +
+				" out of range"}
+	}
+
 	if err := config.SessionBus.CheckInterfaces("session"); err != nil {
 		return err
 	}

hst/config_test.go

@@ -22,6 +22,10 @@ func TestConfigValidate(t *testing.T) {
 			Msg: "identity -1 out of range"}},
 		{"identity upper", &hst.Config{Identity: 10000}, &hst.AppError{Step: "validate configuration", Err: hst.ErrIdentityBounds,
 			Msg: "identity 10000 out of range"}},
+		{"sched lower", &hst.Config{SchedPolicy: -1}, &hst.AppError{Step: "validate configuration", Err: hst.ErrSchedPolicyBounds,
+			Msg: "scheduling policy -1 out of range"}},
+		{"sched upper", &hst.Config{SchedPolicy: 0xcafe}, &hst.AppError{Step: "validate configuration", Err: hst.ErrSchedPolicyBounds,
+			Msg: "scheduling policy 51966 out of range"}},
 		{"dbus session", &hst.Config{SessionBus: &hst.BusConfig{See: []string{""}}},
 			&hst.BadInterfaceError{Interface: "", Segment: "session"}},
 		{"dbus system", &hst.Config{SystemBus: &hst.BusConfig{See: []string{""}}},

hst/container.go

@@ -16,18 +16,20 @@ const PrivateTmp = "/.hakurei"
 var AbsPrivateTmp = check.MustAbs(PrivateTmp)
 
 const (
-	// WaitDelayDefault is used when WaitDelay has its zero value.
+	// WaitDelayDefault is used when WaitDelay has the zero value.
 	WaitDelayDefault = 5 * time.Second
-	// WaitDelayMax is used if WaitDelay exceeds its value.
+	// WaitDelayMax is used when WaitDelay exceeds its value.
 	WaitDelayMax = 30 * time.Second
 )
 
 const (
 	// ExitFailure is returned if the container fails to start.
 	ExitFailure = iota + 1
-	// ExitCancel is returned if the container is terminated by a shim-directed signal which cancels its context.
+	// ExitCancel is returned if the container is terminated by a shim-directed
+	// signal which cancels its context.
 	ExitCancel
-	// ExitOrphan is returned when the shim is orphaned before priv side delivers a signal.
+	// ExitOrphan is returned when the shim is orphaned before priv side process
+	// delivers a signal.
 	ExitOrphan
 
 	// ExitRequest is returned when the priv side process requests shim exit.
@@ -38,10 +40,12 @@ const (
 type Flags uintptr
 
 const (
-	// FMultiarch unblocks syscalls required for multiarch to work on applicable targets.
+	// FMultiarch unblocks system calls required for multiarch to work on
+	// multiarch-enabled targets (amd64, arm64).
 	FMultiarch Flags = 1 << iota
 
-	// FSeccompCompat changes emitted seccomp filter programs to be identical to that of Flatpak.
+	// FSeccompCompat changes emitted seccomp filter programs to be identical to
+	// that of Flatpak in enabled rulesets.
 	FSeccompCompat
 	// FDevel unblocks ptrace and friends.
 	FDevel
@@ -54,12 +58,15 @@ const (
 	// FTty unblocks dangerous terminal I/O (faking input).
 	FTty
 
-	// FMapRealUID maps the target user uid to the privileged user uid in the container user namespace.
-	//	Some programs fail to connect to dbus session running as a different uid,
-	//	this option works around it by mapping priv-side caller uid in container.
+	// FMapRealUID maps the target user uid to the privileged user uid in the
+	// container user namespace.
+	//
+	// Some programs fail to connect to dbus session running as a different uid,
+	// this option works around it by mapping priv-side caller uid in container.
 	FMapRealUID
 
-	// FDevice mount /dev/ from the init mount namespace as-is in the container mount namespace.
+	// FDevice mount /dev/ from the init mount namespace as is in the container
+	// mount namespace.
 	FDevice
 
 	// FShareRuntime shares XDG_RUNTIME_DIR between containers under the same identity.
@@ -112,30 +119,37 @@ func (flags Flags) String() string {
 	}
 }
 
-// ContainerConfig describes the container configuration to be applied to an underlying [container].
+// ContainerConfig describes the container configuration to be applied to an
+// underlying [container]. It is validated by [Config.Validate].
 type ContainerConfig struct {
 	// Container UTS namespace hostname.
 	Hostname string `json:"hostname,omitempty"`
 
 	// Duration in nanoseconds to wait for after interrupting the initial process.
-	// Defaults to [WaitDelayDefault] if zero, or [WaitDelayMax] if greater than [WaitDelayMax].
-	// Values lesser than zero is equivalent to zero, bypassing [WaitDelayDefault].
+	//
+	// Defaults to [WaitDelayDefault] if zero, or [WaitDelayMax] if greater than
+	// [WaitDelayMax]. Values lesser than zero is equivalent to zero, bypassing
+	// [WaitDelayDefault].
 	WaitDelay time.Duration `json:"wait_delay,omitempty"`
 
 	// Initial process environment variables.
 	Env map[string]string `json:"env"`
 
-	/* Container mount points.
-
-	If the first element targets /, it is inserted early and excluded from path hiding. */
+	// Container mount points.
+	//
+	// If the first element targets /, it is inserted early and excluded from
+	// path hiding. Otherwise, an anonymous instance of tmpfs is set up on /.
 	Filesystem []FilesystemConfigJSON `json:"filesystem"`
 
-	// String used as the username of the emulated user, validated against the default NAME_REGEX from adduser.
+	// String used as the username of the emulated user, validated against the
+	// default NAME_REGEX from adduser.
+	//
 	// Defaults to passwd name of target uid or chronos.
 	Username string `json:"username,omitempty"`
 	// Pathname of shell in the container filesystem to use for the emulated user.
 	Shell *check.Absolute `json:"shell"`
-	// Directory in the container filesystem to enter and use as the home directory of the emulated user.
+	// Directory in the container filesystem to enter and use as the home
+	// directory of the emulated user.
 	Home *check.Absolute `json:"home"`
 
 	// Pathname to executable file in the container filesystem.
@@ -148,6 +162,7 @@ type ContainerConfig struct {
 }
 
 // ContainerConfigF is [ContainerConfig] stripped of its methods.
+//
 // The [ContainerConfig.Flags] field does not survive a [json] round trip.
 type ContainerConfigF ContainerConfig
 

hst/dbus.go

@@ -5,8 +5,26 @@ import (
 	"strings"
 )
 
-// BadInterfaceError is returned when Interface fails an undocumented check in xdg-dbus-proxy,
-// which would have cause a silent failure.
+// BadInterfaceError is returned when Interface fails an undocumented check in
+// xdg-dbus-proxy, which would have cause a silent failure.
+//
+// xdg-dbus-proxy fails without output when this condition is not met:
+//
+//	char *dot = strrchr (filter->interface, '.');
+//	if (dot != NULL)
+//	  {
+//	    *dot = 0;
+//	    if (strcmp (dot + 1, "*") != 0)
+//	      filter->member = g_strdup (dot + 1);
+//	  }
+//
+// trim ".*" since they are removed before searching for '.':
+//
+//	if (g_str_has_suffix (name, ".*"))
+//	  {
+//	    name[strlen (name) - 2] = 0;
+//	    wildcard = TRUE;
+//	  }
 type BadInterfaceError struct {
 	// Interface is the offending interface string.
 	Interface string
@@ -19,7 +37,8 @@ func (e *BadInterfaceError) Error() string {
 	if e == nil {
 		return "<nil>"
 	}
-	return "bad interface string " + strconv.Quote(e.Interface) + " in " + e.Segment + " bus configuration"
+	return "bad interface string " + strconv.Quote(e.Interface) +
+		" in " + e.Segment + " bus configuration"
 }
 
 // BusConfig configures the xdg-dbus-proxy process.
@@ -76,31 +95,14 @@ func (c *BusConfig) Interfaces(yield func(string) bool) {
 	}
 }
 
-// CheckInterfaces checks for invalid interface strings based on an undocumented check in xdg-dbus-error,
-// returning [BadInterfaceError] if one is encountered.
+// CheckInterfaces checks for invalid interface strings based on an undocumented
+// check in xdg-dbus-error, returning [BadInterfaceError] if one is encountered.
 func (c *BusConfig) CheckInterfaces(segment string) error {
 	if c == nil {
 		return nil
 	}
 
 	for iface := range c.Interfaces {
-		/*
-			xdg-dbus-proxy fails without output when this condition is not met:
-				char *dot = strrchr (filter->interface, '.');
-				if (dot != NULL)
-				  {
-				    *dot = 0;
-				    if (strcmp (dot + 1, "*") != 0)
-				      filter->member = g_strdup (dot + 1);
-				  }
-
-			trim ".*" since they are removed before searching for '.':
-				if (g_str_has_suffix (name, ".*"))
-				  {
-				    name[strlen (name) - 2] = 0;
-				    wildcard = TRUE;
-				  }
-		*/
 		if strings.IndexByte(strings.TrimSuffix(iface, ".*"), '.') == -1 {
 			return &BadInterfaceError{iface, segment}
 		}

hst/enablement.go

@@ -11,15 +11,17 @@ import (
 type Enablement byte
 
 const (
-	// EWayland exposes a wayland pathname socket via security-context-v1.
+	// EWayland exposes a Wayland pathname socket via security-context-v1.
 	EWayland Enablement = 1 << iota
-	// EX11 adds the target user via X11 ChangeHosts and exposes the X11 pathname socket.
+	// EX11 adds the target user via X11 ChangeHosts and exposes the X11
+	// pathname socket.
 	EX11
 	// EDBus enables the per-container xdg-dbus-proxy daemon.
 	EDBus
 	// EPipeWire exposes a pipewire pathname socket via SecurityContext.
 	EPipeWire
-	// EPulse copies the PulseAudio cookie to [hst.PrivateTmp] and exposes the PulseAudio socket.
+	// EPulse copies the PulseAudio cookie to [hst.PrivateTmp] and exposes the
+	// PulseAudio socket.
 	EPulse
 
 	// EM is a noop.

hst/fs.go

@@ -24,7 +24,8 @@ type FilesystemConfig interface {
 	fmt.Stringer
 }
 
-// The Ops interface enables [FilesystemConfig] to queue container ops without depending on the container package.
+// The Ops interface enables [FilesystemConfig] to queue container ops without
+// depending on the container package.
 type Ops interface {
 	// Tmpfs appends an op that mounts tmpfs on a container path.
 	Tmpfs(target *check.Absolute, size int, perm os.FileMode) Ops
@@ -41,12 +42,15 @@ type Ops interface {
 	// Link appends an op that creates a symlink in the container filesystem.
 	Link(target *check.Absolute, linkName string, dereference bool) Ops
 
-	// Root appends an op that expands a directory into a toplevel bind mount mirror on container root.
+	// Root appends an op that expands a directory into a toplevel bind mount
+	// mirror on container root.
 	Root(host *check.Absolute, flags int) Ops
-	// Etc appends an op that expands host /etc into a toplevel symlink mirror with /etc semantics.
+	// Etc appends an op that expands host /etc into a toplevel symlink mirror
+	// with /etc semantics.
 	Etc(host *check.Absolute, prefix string) Ops
 
-	// Daemon appends an op that starts a daemon in the container and blocks until target appears.
+	// Daemon appends an op that starts a daemon in the container and blocks
+	// until target appears.
 	Daemon(target, path *check.Absolute, args ...string) Ops
 }
 
@@ -61,7 +65,8 @@ type ApplyState struct {
 // ErrFSNull is returned by [json] on encountering a null [FilesystemConfig] value.
 var ErrFSNull = errors.New("unexpected null in mount point")
 
-// FSTypeError is returned when [ContainerConfig.Filesystem] contains an entry with invalid type.
+// FSTypeError is returned when [ContainerConfig.Filesystem] contains an entry
+// with invalid type.
 type FSTypeError string
 
 func (f FSTypeError) Error() string { return fmt.Sprintf("invalid filesystem type %q", string(f)) }

hst/fslink.go

@@ -18,7 +18,9 @@ type FSLink struct {
 	Target *check.Absolute `json:"dst"`
 	// Arbitrary linkname value store in the symlink.
 	Linkname string `json:"linkname"`
-	// Whether to treat Linkname as an absolute pathname and dereference before creating the link.
+
+	// Whether to treat Linkname as an absolute pathname and dereference before
+	// creating the link.
 	Dereference bool `json:"dereference,omitempty"`
 }
 

hst/fsoverlay.go

@@ -19,9 +19,11 @@ type FSOverlay struct {
 
 	// Any filesystem, does not need to be on a writable filesystem, must not be nil.
 	Lower []*check.Absolute `json:"lower"`
-	// The upperdir is normally on a writable filesystem, leave as nil to mount Lower readonly.
+	// The upperdir is normally on a writable filesystem, leave as nil to mount
+	// Lower readonly.
 	Upper *check.Absolute `json:"upper,omitempty"`
-	// The workdir needs to be an empty directory on the same filesystem as Upper, must not be nil if Upper is populated.
+	// The workdir needs to be an empty directory on the same filesystem as
+	// Upper, must not be nil if Upper is populated.
 	Work *check.Absolute `json:"work,omitempty"`
 }
 

hst/hst.go

@@ -44,11 +44,13 @@ func (e *AppError) Message() string {
 type Paths struct {
 	// Temporary directory returned by [os.TempDir], usually equivalent to [fhs.AbsTmp].
 	TempDir *check.Absolute `json:"temp_dir"`
-	// Shared directory specific to the hsu userid, usually (`/tmp/hakurei.%d`, [Info.User]).
+	// Shared directory specific to the hsu userid, usually
+	// (`/tmp/hakurei.%d`, [Info.User]).
 	SharePath *check.Absolute `json:"share_path"`
 	// Checked XDG_RUNTIME_DIR value, usually (`/run/user/%d`, uid).
 	RuntimePath *check.Absolute `json:"runtime_path"`
-	// Shared directory specific to the hsu userid located in RuntimePath, usually (`/run/user/%d/hakurei`, uid).
+	// Shared directory specific to the hsu userid located in RuntimePath,
+	// usually (`/run/user/%d/hakurei`, uid).
 	RunDirPath *check.Absolute `json:"run_dir_path"`
 }
 
@@ -74,10 +76,23 @@ func Template() *Config {
 
 		SessionBus: &BusConfig{
 			See: nil,
-			Talk: []string{"org.freedesktop.Notifications", "org.freedesktop.FileManager1", "org.freedesktop.ScreenSaver",
-				"org.freedesktop.secrets", "org.kde.kwalletd5", "org.kde.kwalletd6", "org.gnome.SessionManager"},
-			Own: []string{"org.chromium.Chromium.*", "org.mpris.MediaPlayer2.org.chromium.Chromium.*",
-				"org.mpris.MediaPlayer2.chromium.*"},
+			Talk: []string{
+				"org.freedesktop.Notifications",
+				"org.freedesktop.FileManager1",
+				"org.freedesktop.ScreenSaver",
+				"org.freedesktop.secrets",
+
+				"org.kde.kwalletd5",
+				"org.kde.kwalletd6",
+
+				"org.gnome.SessionManager",
+			},
+			Own: []string{
+				"org.chromium.Chromium.*",
+
+				"org.mpris.MediaPlayer2.org.chromium.Chromium.*",
+				"org.mpris.MediaPlayer2.chromium.*",
+			},
 			Call:      map[string]string{"org.freedesktop.portal.*": "*"},
 			Broadcast: map[string]string{"org.freedesktop.portal.*": "@/org/freedesktop/portal/*"},
 			Log:       false,
@@ -112,7 +127,12 @@ func Template() *Config {
 				"GOOGLE_DEFAULT_CLIENT_SECRET": "OTJgUOQcT7lO7GsGZq2G4IlT",
 			},
 			Filesystem: []FilesystemConfigJSON{
-				{&FSBind{Target: fhs.AbsRoot, Source: fhs.AbsVarLib.Append("hakurei/base/org.debian"), Write: true, Special: true}},
+				{&FSBind{
+					Target:  fhs.AbsRoot,
+					Source:  fhs.AbsVarLib.Append("hakurei/base/org.debian"),
+					Write:   true,
+					Special: true,
+				}},
 				{&FSBind{Target: fhs.AbsEtc, Source: fhs.AbsEtc, Special: true}},
 				{&FSEphemeral{Target: fhs.AbsTmp, Write: true, Perm: 0755}},
 				{&FSOverlay{
@@ -121,11 +141,27 @@ func Template() *Config {
 					Upper:  fhs.AbsVarLib.Append("hakurei/nix/u0/org.chromium.Chromium/rw-store/upper"),
 					Work:   fhs.AbsVarLib.Append("hakurei/nix/u0/org.chromium.Chromium/rw-store/work"),
 				}},
-				{&FSLink{Target: fhs.AbsRun.Append("current-system"), Linkname: "/run/current-system", Dereference: true}},
-				{&FSLink{Target: fhs.AbsRun.Append("opengl-driver"), Linkname: "/run/opengl-driver", Dereference: true}},
-				{&FSBind{Source: fhs.AbsVarLib.Append("hakurei/u0/org.chromium.Chromium"),
-					Target: check.MustAbs("/data/data/org.chromium.Chromium"), Write: true, Ensure: true}},
-				{&FSBind{Source: fhs.AbsDev.Append("dri"), Device: true, Optional: true}},
+				{&FSLink{
+					Target:      fhs.AbsRun.Append("current-system"),
+					Linkname:    "/run/current-system",
+					Dereference: true,
+				}},
+				{&FSLink{
+					Target:      fhs.AbsRun.Append("opengl-driver"),
+					Linkname:    "/run/opengl-driver",
+					Dereference: true,
+				}},
+				{&FSBind{
+					Source: fhs.AbsVarLib.Append("hakurei/u0/org.chromium.Chromium"),
+					Target: check.MustAbs("/data/data/org.chromium.Chromium"),
+					Write:  true,
+					Ensure: true,
+				}},
+				{&FSBind{
+					Source:   fhs.AbsDev.Append("dri"),
+					Device:   true,
+					Optional: true,
+				}},
 			},
 
 			Username: "chronos",

hst/instance.go

@@ -12,10 +12,12 @@ import (
 // An ID is a unique identifier held by a running hakurei container.
 type ID [16]byte
 
-// ErrIdentifierLength is returned when encountering a [hex] representation of [ID] with unexpected length.
+// ErrIdentifierLength is returned when encountering a [hex] representation of
+// [ID] with unexpected length.
 var ErrIdentifierLength = errors.New("identifier string has unexpected length")
 
-// IdentifierDecodeError is returned by [ID.UnmarshalText] to provide relevant error descriptions.
+// IdentifierDecodeError is returned by [ID.UnmarshalText] to provide relevant
+// error descriptions.
 type IdentifierDecodeError struct{ Err error }
 
 func (e IdentifierDecodeError) Unwrap() error { return e.Err }
@@ -23,7 +25,10 @@ func (e IdentifierDecodeError) Error() string {
 	var invalidByteError hex.InvalidByteError
 	switch {
 	case errors.As(e.Err, &invalidByteError):
-		return fmt.Sprintf("got invalid byte %#U in identifier", rune(invalidByteError))
+		return fmt.Sprintf(
+			"got invalid byte %#U in identifier",
+			rune(invalidByteError),
+		)
 	case errors.Is(e.Err, hex.ErrLength):
 		return "odd length identifier hex string"
 
@@ -41,7 +46,9 @@ func (a *ID) CreationTime() time.Time {
 }
 
 // NewInstanceID creates a new unique [ID].
-func NewInstanceID(id *ID) error { return newInstanceID(id, uint64(time.Now().UnixNano())) }
+func NewInstanceID(id *ID) error {
+	return newInstanceID(id, uint64(time.Now().UnixNano()))
+}
 
 // newInstanceID creates a new unique [ID] with the specified timestamp.
 func newInstanceID(id *ID, p uint64) error {

internal/outcome/hsu.go

@@ -38,6 +38,7 @@ func (h *Hsu) ensureDispatcher() {
 }
 
 // ID returns the current user hsurc identifier.
+//
 // [ErrHsuAccess] is returned if the current user is not in hsurc.
 func (h *Hsu) ID() (int, error) {
 	h.ensureDispatcher()

internal/outcome/outcome.go

@@ -1,4 +1,5 @@
-// Package outcome implements the outcome of the privileged and container sides of a hakurei container.
+// Package outcome implements the outcome of the privileged and container sides
+// of a hakurei container.
 package outcome
 
 import (
@@ -27,8 +28,9 @@ func Info() *hst.Info {
 	return &hi
 }
 
-// envAllocSize is the initial size of the env map pre-allocated when the configured env map is nil.
-// It should be large enough to fit all insertions by outcomeOp.toContainer.
+// envAllocSize is the initial size of the env map pre-allocated when the
+// configured env map is nil. It should be large enough to fit all insertions by
+// outcomeOp.toContainer.
 const envAllocSize = 1 << 6
 
 func newInt(v int) *stringPair[int] { return &stringPair[int]{v, strconv.Itoa(v)} }
@@ -43,7 +45,8 @@ func (s *stringPair[T]) unwrap() T      { return s.v }
 func (s *stringPair[T]) String() string { return s.s }
 
 // outcomeState is copied to the shim process and available while applying outcomeOp.
-// This is transmitted from the priv side to the shim, so exported fields should be kept to a minimum.
+// This is transmitted from the priv side to the shim, so exported fields should
+// be kept to a minimum.
 type outcomeState struct {
 	// Params only used by the shim process. Populated by populateEarly.
 	Shim *shimParams
@@ -89,14 +92,25 @@ func (s *outcomeState) valid() bool {
 		s.Paths != nil
 }
 
-// newOutcomeState returns the address of a new outcomeState with its exported fields populated via syscallDispatcher.
+// newOutcomeState returns the address of a new outcomeState with its exported
+// fields populated via syscallDispatcher.
 func newOutcomeState(k syscallDispatcher, msg message.Msg, id *hst.ID, config *hst.Config, hsu *Hsu) *outcomeState {
 	s := outcomeState{
-		Shim:      &shimParams{PrivPID: k.getpid(), Verbose: msg.IsVerbose()},
-		ID:        id,
-		Identity:  config.Identity,
-		UserID:    hsu.MustID(msg),
-		Paths:     env.CopyPathsFunc(k.fatalf, k.tempdir, func(key string) string { v, _ := k.lookupEnv(key); return v }),
+		Shim: &shimParams{
+			PrivPID: k.getpid(),
+			Verbose: msg.IsVerbose(),
+
+			SchedPolicy:   config.SchedPolicy,
+			SchedPriority: config.SchedPriority,
+		},
+
+		ID:       id,
+		Identity: config.Identity,
+		UserID:   hsu.MustID(msg),
+		Paths: env.CopyPathsFunc(k.fatalf, k.tempdir, func(key string) string {
+			v, _ := k.lookupEnv(key)
+			return v
+		}),
 		Container: config.Container,
 	}
 
@@ -121,6 +135,7 @@ func newOutcomeState(k syscallDispatcher, msg message.Msg, id *hst.ID, config *h
 }
 
 // populateLocal populates unexported fields from transmitted exported fields.
+//
 // These fields are cheaper to recompute per-process.
 func (s *outcomeState) populateLocal(k syscallDispatcher, msg message.Msg) error {
 	if !s.valid() || k == nil || msg == nil {
@@ -136,7 +151,10 @@ func (s *outcomeState) populateLocal(k syscallDispatcher, msg message.Msg) error
 	s.id = &stringPair[hst.ID]{*s.ID, s.ID.String()}
 
 	s.Copy(&s.sc, s.UserID)
-	msg.Verbosef("process share directory at %q, runtime directory at %q", s.sc.SharePath, s.sc.RunDirPath)
+	msg.Verbosef(
+		"process share directory at %q, runtime directory at %q",
+		s.sc.SharePath, s.sc.RunDirPath,
+	)
 
 	s.identity = newInt(s.Identity)
 	s.mapuid, s.mapgid = newInt(s.Mapuid), newInt(s.Mapgid)
@@ -146,17 +164,25 @@ func (s *outcomeState) populateLocal(k syscallDispatcher, msg message.Msg) error
 }
 
 // instancePath returns a path formatted for outcomeStateSys.instance.
+//
 // This method must only be called from outcomeOp.toContainer if
 // outcomeOp.toSystem has already called outcomeStateSys.instance.
-func (s *outcomeState) instancePath() *check.Absolute { return s.sc.SharePath.Append(s.id.String()) }
+func (s *outcomeState) instancePath() *check.Absolute {
+	return s.sc.SharePath.Append(s.id.String())
+}
 
 // runtimePath returns a path formatted for outcomeStateSys.runtime.
+//
 // This method must only be called from outcomeOp.toContainer if
 // outcomeOp.toSystem has already called outcomeStateSys.runtime.
-func (s *outcomeState) runtimePath() *check.Absolute { return s.sc.RunDirPath.Append(s.id.String()) }
+func (s *outcomeState) runtimePath() *check.Absolute {
+	return s.sc.RunDirPath.Append(s.id.String())
+}
 
 // outcomeStateSys wraps outcomeState and [system.I]. Used on the priv side only.
-// Implementations of outcomeOp must not access fields other than sys unless explicitly stated.
+//
+// Implementations of outcomeOp must not access fields other than sys unless
+// explicitly stated.
 type outcomeStateSys struct {
 	// Whether XDG_RUNTIME_DIR is used post hsu.
 	useRuntimeDir bool
@@ -219,6 +245,7 @@ func (state *outcomeStateSys) ensureRuntimeDir() {
 }
 
 // instance returns the pathname to a process-specific directory within TMPDIR.
+//
 // This directory must only hold entries bound to [system.Process].
 func (state *outcomeStateSys) instance() *check.Absolute {
 	if state.sharePath != nil {
@@ -230,6 +257,7 @@ func (state *outcomeStateSys) instance() *check.Absolute {
 }
 
 // runtime returns the pathname to a process-specific directory within XDG_RUNTIME_DIR.
+//
 // This directory must only hold entries bound to [system.Process].
 func (state *outcomeStateSys) runtime() *check.Absolute {
 	if state.runtimeSharePath != nil {
@@ -242,22 +270,29 @@ func (state *outcomeStateSys) runtime() *check.Absolute {
 	return state.runtimeSharePath
 }
 
-// outcomeStateParams wraps outcomeState and [container.Params]. Used on the shim side only.
+// outcomeStateParams wraps outcomeState and [container.Params].
+//
+// Used on the shim side only.
 type outcomeStateParams struct {
-	// Overrides the embedded [container.Params] in [container.Container]. The Env field must not be used.
+	// Overrides the embedded [container.Params] in [container.Container].
+	//
+	// The Env field must not be used.
 	params *container.Params
 	// Collapsed into the Env slice in [container.Params] by the final outcomeOp.
 	env map[string]string
 
-	// Filesystems with the optional root sliced off if present. Populated by spParamsOp.
-	// Safe for use by spFilesystemOp.
+	// Filesystems with the optional root sliced off if present.
+	//
+	// Populated by spParamsOp. Safe for use by spFilesystemOp.
 	filesystem []hst.FilesystemConfigJSON
 
 	// Inner XDG_RUNTIME_DIR default formatting of `/run/user/%d` via mapped uid.
+	//
 	// Populated by spRuntimeOp.
 	runtimeDir *check.Absolute
 
 	// Path to pipewire-pulse server.
+	//
 	// Populated by spPipeWireOp if DirectPipeWire is false.
 	pipewirePulsePath *check.Absolute
 
@@ -265,25 +300,32 @@ type outcomeStateParams struct {
 	*outcomeState
 }
 
-// errNotEnabled is returned by outcomeOp.toSystem and used internally to exclude an outcomeOp from transmission.
+// errNotEnabled is returned by outcomeOp.toSystem and used internally to
+// exclude an outcomeOp from transmission.
 var errNotEnabled = errors.New("op not enabled in the configuration")
 
-// An outcomeOp inflicts an outcome on [system.I] and contains enough information to
-// inflict it on [container.Params] in a separate process.
-// An implementation of outcomeOp must store cross-process states in exported fields only.
+// An outcomeOp inflicts an outcome on [system.I] and contains enough
+// information to inflict it on [container.Params] in a separate process.
+//
+// An implementation of outcomeOp must store cross-process states in exported
+// fields only.
 type outcomeOp interface {
 	// toSystem inflicts the current outcome on [system.I] in the priv side process.
 	toSystem(state *outcomeStateSys) error
 
-	// toContainer inflicts the current outcome on [container.Params] in the shim process.
-	// The implementation must not write to the Env field of [container.Params] as it will be overwritten
-	// by flattened env map.
+	// toContainer inflicts the current outcome on [container.Params] in the
+	// shim process.
+	//
+	// Implementations must not write to the Env field of [container.Params]
+	// as it will be overwritten by flattened env map.
 	toContainer(state *outcomeStateParams) error
 }
 
-// toSystem calls the outcomeOp.toSystem method on all outcomeOp implementations and populates shimParams.Ops.
-// This function assumes the caller has already called the Validate method on [hst.Config]
-// and checked that it returns nil.
+// toSystem calls the outcomeOp.toSystem method on all outcomeOp implementations
+// and populates shimParams.Ops.
+//
+// This function assumes the caller has already called the Validate method on
+// [hst.Config] and checked that it returns nil.
 func (state *outcomeStateSys) toSystem() error {
 	if state.Shim == nil || state.Shim.Ops != nil {
 		return newWithMessage("invalid ops state reached")

internal/outcome/process.go

@@ -30,7 +30,9 @@ const (
 )
 
 // NewStore returns the address of a new instance of [store.Store].
-func NewStore(sc *hst.Paths) *store.Store { return store.New(sc.SharePath.Append("state")) }
+func NewStore(sc *hst.Paths) *store.Store {
+	return store.New(sc.SharePath.Append("state"))
+}
 
 // main carries out outcome and terminates. main does not return.
 func (k *outcome) main(msg message.Msg, identifierFd int) {
@@ -116,7 +118,11 @@ func (k *outcome) main(msg message.Msg, identifierFd int) {
 		processStatePrev, processStateCur = processStateCur, processState
 
 		if !processTime.IsZero() && processStatePrev != processLifecycle {
-			msg.Verbosef("state %d took %.2f ms", processStatePrev, float64(time.Since(processTime).Nanoseconds())/1e6)
+			msg.Verbosef(
+				"state %d took %.2f ms",
+				processStatePrev,
+				float64(time.Since(processTime).Nanoseconds())/1e6,
+			)
 		}
 		processTime = time.Now()
 
@@ -141,7 +147,10 @@ func (k *outcome) main(msg message.Msg, identifierFd int) {
 
 		case processCommit:
 			if isBeforeRevert {
-				perrorFatal(newWithMessage("invalid transition to commit state"), "commit", processLifecycle)
+				perrorFatal(
+					newWithMessage("invalid transition to commit state"),
+					"commit", processLifecycle,
+				)
 				continue
 			}
 
@@ -238,15 +247,26 @@ func (k *outcome) main(msg message.Msg, identifierFd int) {
 
 			case <-func() chan struct{} {
 				w := make(chan struct{})
-				// this ties processLifecycle to ctx with the additional compensated timeout duration
-				// to allow transition to the next state on a locked up shim
-				go func() { <-ctx.Done(); time.Sleep(k.state.Shim.WaitDelay + shimWaitTimeout); close(w) }()
+				// This ties processLifecycle to ctx with the additional
+				// compensated timeout duration to allow transition to the next
+				// state on a locked up shim.
+				go func() {
+					<-ctx.Done()
+					time.Sleep(k.state.Shim.WaitDelay + shimWaitTimeout)
+					close(w)
+				}()
 				return w
 			}():
-				// this is only reachable when wait did not return within shimWaitTimeout, after its WaitDelay has elapsed.
-				// This is different from the container failing to terminate within its timeout period, as that is enforced
-				// by the shim. This path is instead reached when there is a lockup in shim preventing it from completing.
-				msg.GetLogger().Printf("process %d did not terminate", shimCmd.Process.Pid)
+				// This is only reachable when wait did not return within
+				// shimWaitTimeout, after its WaitDelay has elapsed. This is
+				// different from the container failing to terminate within its
+				// timeout period, as that is enforced by the shim. This path is
+				// instead reached when there is a lockup in shim preventing it
+				// from completing.
+				msg.GetLogger().Printf(
+					"process %d did not terminate",
+					shimCmd.Process.Pid,
+				)
 			}
 			msg.Resume()
 
@@ -271,8 +291,8 @@ func (k *outcome) main(msg message.Msg, identifierFd int) {
 				ec := system.Process
 
 				if entries, _, err := handle.Entries(); err != nil {
-					// it is impossible to continue from this point,
-					// per-process state will be reverted to limit damage
+					// it is impossible to continue from this point, per-process
+					// state will be reverted to limit damage
 					perror(err, "read store segment entries")
 				} else {
 					// accumulate enablements of remaining instances
@@ -295,7 +315,10 @@ func (k *outcome) main(msg message.Msg, identifierFd int) {
 					if n == 0 {
 						ec |= system.User
 					} else {
-						msg.Verbosef("found %d instances, cleaning up without user-scoped operations", n)
+						msg.Verbosef(
+							"found %d instances, cleaning up without user-scoped operations",
+							n,
+						)
 					}
 					ec |= rt ^ (hst.EWayland | hst.EX11 | hst.EDBus | hst.EPulse)
 					if msg.IsVerbose() {
@@ -335,7 +358,9 @@ func (k *outcome) main(msg message.Msg, identifierFd int) {
 
 // start starts the shim via cmd/hsu.
 //
-// If successful, a [time.Time] value for [hst.State] is stored in the value pointed to by startTime.
+// If successful, a [time.Time] value for [hst.State] is stored in the value
+// pointed to by startTime.
+//
 // The resulting [exec.Cmd] and write end of the shim setup pipe is returned.
 func (k *outcome) start(ctx context.Context, msg message.Msg,
 	hsuPath *check.Absolute,

internal/outcome/shim.go

@@ -37,9 +37,12 @@ const (
 	shimMsgBadPID = C.HAKUREI_SHIM_BAD_PID
 )
 
-// setupContSignal sets up the SIGCONT signal handler for the cross-uid shim exit hack.
-// The signal handler is implemented in C, signals can be processed by reading from the returned reader.
-// The returned function must be called after all signal processing concludes.
+// setupContSignal sets up the SIGCONT signal handler for the cross-uid shim
+// exit hack.
+//
+// The signal handler is implemented in C, signals can be processed by reading
+// from the returned reader. The returned function must be called after all
+// signal processing concludes.
 func setupContSignal(pid int) (io.ReadCloser, func(), error) {
 	if r, w, err := os.Pipe(); err != nil {
 		return nil, nil, err
@@ -51,22 +54,30 @@ func setupContSignal(pid int) (io.ReadCloser, func(), error) {
 	}
 }
 
-// shimEnv is the name of the environment variable storing decimal representation of
-// setup pipe fd for [container.Receive].
+// shimEnv is the name of the environment variable storing decimal representation
+// of setup pipe fd for [container.Receive].
 const shimEnv = "HAKUREI_SHIM"
 
 // shimParams is embedded in outcomeState and transmitted from priv side to shim.
 type shimParams struct {
-	// Priv side pid, checked against ppid in signal handler for the syscall.SIGCONT hack.
+	// Priv side pid, checked against ppid in signal handler for the
+	// syscall.SIGCONT hack.
 	PrivPID int
 
-	// Duration to wait for after the initial process receives os.Interrupt before the container is killed.
+	// Duration to wait for after the initial process receives os.Interrupt
+	// before the container is killed.
+	//
 	// Limits are enforced on the priv side.
 	WaitDelay time.Duration
 
 	// Verbosity pass through from [message.Msg].
 	Verbose bool
 
+	// Copied from [hst.Config].
+	SchedPolicy std.SchedPolicy
+	// Copied from [hst.Config].
+	SchedPriority std.Int
+
 	// Outcome setup ops, contains setup state. Populated by outcome.finalise.
 	Ops []outcomeOp
 }
@@ -77,7 +88,9 @@ func (p *shimParams) valid() bool { return p != nil && p.PrivPID > 0 }
 // shimName is the prefix used by log.std in the shim process.
 const shimName = "shim"
 
-// Shim is called by the main function of the shim process and runs as the unconstrained target user.
+// Shim is called by the main function of the shim process and runs as the
+// unconstrained target user.
+//
 // Shim does not return.
 func Shim(msg message.Msg) {
 	if msg == nil {
@@ -131,7 +144,8 @@ func (sp *shimPrivate) destroy() {
 }
 
 const (
-	// shimPipeWireTimeout is the duration pipewire-pulse is allowed to run before its socket becomes available.
+	// shimPipeWireTimeout is the duration pipewire-pulse is allowed to run
+	// before its socket becomes available.
 	shimPipeWireTimeout = 5 * time.Second
 )
 
@@ -262,6 +276,9 @@ func shimEntrypoint(k syscallDispatcher) {
 	cancelContainer.Store(&stop)
 	sp := shimPrivate{k: k, id: state.id}
 	z := container.New(ctx, msg)
+	z.SetScheduler = state.Shim.SchedPolicy > 0
+	z.SchedPolicy = state.Shim.SchedPolicy
+	z.SchedPriority = state.Shim.SchedPriority
 	z.Params = *stateParams.params
 	z.Stdin, z.Stdout, z.Stderr = os.Stdin, os.Stdout, os.Stderr
 

internal/outcome/spcontainer.go

@@ -27,7 +27,9 @@ const varRunNscd = fhs.Var + "run/nscd"
 
 func init() { gob.Register(new(spParamsOp)) }
 
-// spParamsOp initialises unordered fields of [container.Params] and the optional root filesystem.
+// spParamsOp initialises unordered fields of [container.Params] and the
+// optional root filesystem.
+//
 // This outcomeOp is hardcoded to always run first.
 type spParamsOp struct {
 	// Value of $TERM, stored during toSystem.
@@ -67,8 +69,8 @@ func (s *spParamsOp) toContainer(state *outcomeStateParams) error {
 		state.params.Args = state.Container.Args
 	}
 
-	// the container is canceled when shim is requested to exit or receives an interrupt or termination signal;
-	// this behaviour is implemented in the shim
+	// The container is cancelled when shim is requested to exit or receives an
+	// interrupt or termination signal. This behaviour is implemented in the shim.
 	state.params.ForwardCancel = state.Shim.WaitDelay > 0
 
 	if state.Container.Flags&hst.FMultiarch != 0 {
@@ -115,7 +117,8 @@ func (s *spParamsOp) toContainer(state *outcomeStateParams) error {
 	} else {
 		state.params.Bind(fhs.AbsDev, fhs.AbsDev, std.BindWritable|std.BindDevice)
 	}
-	// /dev is mounted readonly later on, this prevents /dev/shm from going readonly with it
+	// /dev is mounted readonly later on, this prevents /dev/shm from going
+	// readonly with it
 	state.params.Tmpfs(fhs.AbsDevShm, 0, 01777)
 
 	return nil
@@ -123,7 +126,9 @@ func (s *spParamsOp) toContainer(state *outcomeStateParams) error {
 
 func init() { gob.Register(new(spFilesystemOp)) }
 
-// spFilesystemOp applies configured filesystems to [container.Params], excluding the optional root filesystem.
+// spFilesystemOp applies configured filesystems to [container.Params],
+// excluding the optional root filesystem.
+//
 // This outcomeOp is hardcoded to always run last.
 type spFilesystemOp struct {
 	// Matched paths to cover. Stored during toSystem.
@@ -297,8 +302,8 @@ func (s *spFilesystemOp) toContainer(state *outcomeStateParams) error {
 	return nil
 }
 
-// resolveRoot handles the root filesystem special case for [hst.FilesystemConfig] and additionally resolves autoroot
-// as it requires special handling during path hiding.
+// resolveRoot handles the root filesystem special case for [hst.FilesystemConfig]
+// and additionally resolves autoroot as it requires special handling during path hiding.
 func resolveRoot(c *hst.ContainerConfig) (rootfs hst.FilesystemConfig, filesystem []hst.FilesystemConfigJSON, autoroot *hst.FSBind) {
 	// root filesystem special case
 	filesystem = c.Filesystem
@@ -316,7 +321,8 @@ func resolveRoot(c *hst.ContainerConfig) (rootfs hst.FilesystemConfig, filesyste
 	return
 }
 
-// evalSymlinks calls syscallDispatcher.evalSymlinks but discards errors unwrapping to [fs.ErrNotExist].
+// evalSymlinks calls syscallDispatcher.evalSymlinks but discards errors
+// unwrapping to [fs.ErrNotExist].
 func evalSymlinks(msg message.Msg, k syscallDispatcher, v *string) error {
 	if p, err := k.evalSymlinks(*v); err != nil {
 		if !errors.Is(err, fs.ErrNotExist) {

internal/outcome/spdbus.go

@@ -12,6 +12,7 @@ import (
 func init() { gob.Register(new(spDBusOp)) }
 
 // spDBusOp maintains an xdg-dbus-proxy instance for the container.
+//
 // Runs after spRuntimeOp.
 type spDBusOp struct {
 	// Whether to bind the system bus socket. Populated during toSystem.

internal/outcome/sppipewire.go

@@ -13,9 +13,12 @@ const pipewirePulseName = "pipewire-pulse"
 func init() { gob.Register(new(spPipeWireOp)) }
 
 // spPipeWireOp exports the PipeWire server to the container via SecurityContext.
+//
 // Runs after spRuntimeOp.
 type spPipeWireOp struct {
-	// Path to pipewire-pulse server. Populated during toSystem if DirectPipeWire is false.
+	// Path to pipewire-pulse server.
+	//
+	// Populated during toSystem if DirectPipeWire is false.
 	CompatServerPath *check.Absolute
 }
 

internal/outcome/sppulse.go

@@ -20,6 +20,7 @@ const pulseCookieSizeMax = 1 << 8
 func init() { gob.Register(new(spPulseOp)) }
 
 // spPulseOp exports the PulseAudio server to the container.
+//
 // Runs after spRuntimeOp.
 type spPulseOp struct {
 	// PulseAudio cookie data, populated during toSystem if a cookie is present.
@@ -37,24 +38,40 @@ func (s *spPulseOp) toSystem(state *outcomeStateSys) error {
 
 	if _, err := state.k.stat(pulseRuntimeDir.String()); err != nil {
 		if !errors.Is(err, fs.ErrNotExist) {
-			return &hst.AppError{Step: fmt.Sprintf("access PulseAudio directory %q", pulseRuntimeDir), Err: err}
+			return &hst.AppError{Step: fmt.Sprintf(
+				"access PulseAudio directory %q",
+				pulseRuntimeDir,
+			), Err: err}
 		}
-		return newWithMessageError(fmt.Sprintf("PulseAudio directory %q not found", pulseRuntimeDir), err)
+		return newWithMessageError(fmt.Sprintf(
+			"PulseAudio directory %q not found",
+			pulseRuntimeDir,
+		), err)
 	}
 
 	if fi, err := state.k.stat(pulseSocket.String()); err != nil {
 		if !errors.Is(err, fs.ErrNotExist) {
-			return &hst.AppError{Step: fmt.Sprintf("access PulseAudio socket %q", pulseSocket), Err: err}
+			return &hst.AppError{Step: fmt.Sprintf(
+				"access PulseAudio socket %q",
+				pulseSocket,
+			), Err: err}
 		}
-		return newWithMessageError(fmt.Sprintf("PulseAudio directory %q found but socket does not exist", pulseRuntimeDir), err)
+		return newWithMessageError(fmt.Sprintf(
+			"PulseAudio directory %q found but socket does not exist",
+			pulseRuntimeDir,
+		), err)
 	} else {
 		if m := fi.Mode(); m&0o006 != 0o006 {
-			return newWithMessage(fmt.Sprintf("unexpected permissions on %q: %s", pulseSocket, m))
+			return newWithMessage(fmt.Sprintf(
+				"unexpected permissions on %q: %s",
+				pulseSocket, m,
+			))
 		}
 	}
 
-	// pulse socket is world writable and its parent directory DAC permissions prevents access;
-	// hard link to target-executable share directory to grant access
+	// PulseAudio socket is world writable and its parent directory DAC
+	// permissions prevents access. Hard link to target-executable share
+	// directory to grant access
 	state.sys.Link(pulseSocket, state.runtime().Append("pulse"))
 
 	// load up to pulseCookieSizeMax bytes of pulse cookie for transmission to shim
@@ -62,7 +79,13 @@ func (s *spPulseOp) toSystem(state *outcomeStateSys) error {
 		return err
 	} else if a != nil {
 		s.Cookie = new([pulseCookieSizeMax]byte)
-		if s.CookieSize, err = loadFile(state.msg, state.k, "PulseAudio cookie", a.String(), s.Cookie[:]); err != nil {
+		if s.CookieSize, err = loadFile(
+			state.msg,
+			state.k,
+			"PulseAudio cookie",
+			a.String(),
+			s.Cookie[:],
+		); err != nil {
 			return err
 		}
 	} else {
@@ -101,8 +124,9 @@ func (s *spPulseOp) commonPaths(state *outcomeState) (pulseRuntimeDir, pulseSock
 	return
 }
 
-// discoverPulseCookie attempts to discover the pathname of the PulseAudio cookie of the current user.
-// If both returned pathname and error are nil, the cookie is likely unavailable and can be silently skipped.
+// discoverPulseCookie attempts to discover the pathname of the PulseAudio
+// cookie of the current user. If both returned pathname and error are nil, the
+// cookie is likely unavailable and can be silently skipped.
 func discoverPulseCookie(k syscallDispatcher) (*check.Absolute, error) {
 	const paLocateStep = "locate PulseAudio cookie"
 
@@ -186,7 +210,10 @@ func loadFile(
 				&os.PathError{Op: "stat", Path: pathname, Err: syscall.ENOMEM},
 			)
 		} else if s < int64(n) {
-			msg.Verbosef("%s at %q is %d bytes shorter than expected", description, pathname, int64(n)-s)
+			msg.Verbosef(
+				"%s at %q is %d bytes shorter than expected",
+				description, pathname, int64(n)-s,
+			)
 		} else {
 			msg.Verbosef("loading %d bytes from %q", n, pathname)
 		}

internal/outcome/spruntime.go

@@ -67,7 +67,9 @@ const (
 
 // spRuntimeOp sets up XDG_RUNTIME_DIR inside the container.
 type spRuntimeOp struct {
-	// SessionType determines the value of envXDGSessionType. Populated during toSystem.
+	// SessionType determines the value of envXDGSessionType.
+	//
+	// Populated during toSystem.
 	SessionType uintptr
 }
 

internal/outcome/spwayland.go

@@ -12,9 +12,12 @@ import (
 func init() { gob.Register(new(spWaylandOp)) }
 
 // spWaylandOp exports the Wayland display server to the container.
+//
 // Runs after spRuntimeOp.
 type spWaylandOp struct {
-	// Path to host wayland socket. Populated during toSystem if DirectWayland is true.
+	// Path to host wayland socket.
+	//
+	// Populated during toSystem if DirectWayland is true.
 	SocketPath *check.Absolute
 }
 

internal/outcome/spx11.go

@@ -50,7 +50,10 @@ func (s *spX11Op) toSystem(state *outcomeStateSys) error {
 	if socketPath != nil {
 		if _, err := state.k.stat(socketPath.String()); err != nil {
 			if !errors.Is(err, fs.ErrNotExist) {
-				return &hst.AppError{Step: fmt.Sprintf("access X11 socket %q", socketPath), Err: err}
+				return &hst.AppError{Step: fmt.Sprintf(
+					"access X11 socket %q",
+					socketPath,
+				), Err: err}
 			}
 		} else {
 			state.sys.UpdatePermType(hst.EX11, socketPath, acl.Read, acl.Write, acl.Execute)

internal/pkg/exec.go

@@ -39,8 +39,8 @@ type ExecPath struct {
 	W bool
 }
 
-// SchedPolicy is the [container] scheduling policy.
-var SchedPolicy int
+// SetSchedIdle is whether to set [std.SCHED_IDLE] scheduling priority.
+var SetSchedIdle bool
 
 // PromoteLayers returns artifacts with identical-by-content layers promoted to
 // the highest priority instance, as if mounted via [ExecPath].
@@ -413,7 +413,8 @@ func (a *execArtifact) cure(f *FContext, hostNet bool) (err error) {
 	z.ParentPerm = 0700
 	z.HostNet = hostNet
 	z.Hostname = "cure"
-	z.SchedPolicy = SchedPolicy
+	z.SetScheduler = SetSchedIdle
+	z.SchedPolicy = std.SCHED_IDLE
 	if z.HostNet {
 		z.Hostname = "cure-net"
 	}

internal/rosa/all.go

@@ -19,6 +19,8 @@ const (
 	LLVMRuntimes
 	LLVMClang
 
+	// EarlyInit is the Rosa OS initramfs init program.
+	EarlyInit
 	// ImageInitramfs is the Rosa OS initramfs archive.
 	ImageInitramfs
 
@@ -28,6 +30,8 @@ const (
 	KernelHeaders
 	// KernelSource is a writable kernel source tree installed to [AbsUsrSrc].
 	KernelSource
+	// Firmware is firmware blobs for use with the Linux kernel.
+	Firmware
 
 	ACL
 	ArgpStandalone
@@ -85,9 +89,11 @@ const (
 	NSS
 	NSSCACert
 	Ncurses
+	Nettle
 	Ninja
 	OpenSSL
 	PCRE2
+	Parallel
 	Patch
 	Perl
 	PerlLocaleGettext
@@ -103,12 +109,23 @@ const (
 	PkgConfig
 	Procps
 	Python
+	PythonCfgv
+	PythonDiscovery
+	PythonDistlib
+	PythonFilelock
+	PythonIdentify
 	PythonIniConfig
+	PythonNodeenv
 	PythonPackaging
+	PythonPlatformdirs
 	PythonPluggy
+	PythonPreCommit
 	PythonPyTest
+	PythonPyYAML
 	PythonPygments
+	PythonVirtualenv
 	QEMU
+	Rdfind
 	Rsync
 	Sed
 	Setuptools

internal/rosa/cmake.go

@@ -128,6 +128,9 @@ type CMakeHelper struct {
 	Cache [][2]string
 	// Runs after install.
 	Script string
+
+	// Whether to generate Makefile instead.
+	Make bool
 }
 
 var _ Helper = new(CMakeHelper)
@@ -141,7 +144,10 @@ func (attr *CMakeHelper) name(name, version string) string {
 }
 
 // extra returns a hardcoded slice of [CMake] and [Ninja].
-func (*CMakeHelper) extra(int) []PArtifact {
+func (attr *CMakeHelper) extra(int) []PArtifact {
+	if attr != nil && attr.Make {
+		return []PArtifact{CMake, Make}
+	}
 	return []PArtifact{CMake, Ninja}
 }
 
@@ -173,11 +179,19 @@ func (attr *CMakeHelper) script(name string) string {
 		panic("CACHE must be non-empty")
 	}
 
+	generate := "Ninja"
+	jobs := ""
+	if attr.Make {
+		generate = "'Unix Makefiles'"
+		jobs += ` "--parallel=$(nproc)"`
+	}
+
 	return `
-cmake -G Ninja \
+cmake -G ` + generate + ` \
 	-DCMAKE_C_COMPILER_TARGET="${ROSA_TRIPLE}" \
 	-DCMAKE_CXX_COMPILER_TARGET="${ROSA_TRIPLE}" \
 	-DCMAKE_ASM_COMPILER_TARGET="${ROSA_TRIPLE}" \
+	-DCMAKE_INSTALL_LIBDIR=lib \
 	` + strings.Join(slices.Collect(func(yield func(string) bool) {
 		for _, v := range attr.Cache {
 			if !yield("-D" + v[0] + "=" + v[1]) {
@@ -185,9 +199,9 @@ cmake -G Ninja \
 			}
 		}
 	}), " \\\n\t") + ` \
-	-DCMAKE_INSTALL_PREFIX=/work/system \
+	-DCMAKE_INSTALL_PREFIX=/system \
 	'/usr/src/` + name + `/` + path.Join(attr.Append...) + `'
-cmake --build .
-cmake --install .
+cmake --build .` + jobs + `
+cmake --install . --prefix=/work/system
 ` + attr.Script
 }

internal/rosa/gnu.go

@@ -678,8 +678,8 @@ func init() {
 
 func (t Toolchain) newLibiconv() (pkg.Artifact, string) {
 	const (
-		version  = "1.18"
-		checksum = "iV5q3VxP5VPdJ-X7O5OQI4fGm8VjeYb5viLd1L3eAHg26bbHb2_Qn63XPF3ucVZr"
+		version  = "1.19"
+		checksum = "UibB6E23y4MksNqYmCCrA3zTFO6vJugD1DEDqqWYFZNuBsUWMVMcncb_5pPAr88x"
 	)
 	return t.NewPackage("libiconv", version, pkg.NewHTTPGetTar(
 		nil, "https://ftpmirror.gnu.org/gnu/libiconv/libiconv-"+version+".tar.gz",
@@ -741,6 +741,31 @@ func init() {
 	}
 }
 
+func (t Toolchain) newParallel() (pkg.Artifact, string) {
+	const (
+		version  = "20260222"
+		checksum = "4wxjMi3G2zMxr9hvLcIn6D7_12A3e5UNObeTPhzn7mDAYwsZApmmkxfGPyllQQ7E"
+	)
+	return t.NewPackage("parallel", version, pkg.NewHTTPGetTar(
+		nil, "https://ftpmirror.gnu.org/gnu/parallel/parallel-"+version+".tar.bz2",
+		mustDecode(checksum),
+		pkg.TarBzip2,
+	), nil, (*MakeHelper)(nil),
+		Perl,
+	), version
+}
+func init() {
+	artifactsM[Parallel] = Metadata{
+		f: Toolchain.newParallel,
+
+		Name:        "parallel",
+		Description: "a shell tool for executing jobs in parallel using one or more computers",
+		Website:     "https://www.gnu.org/software/parallel/",
+
+		ID: 5448,
+	}
+}
+
 func (t Toolchain) newBinutils() (pkg.Artifact, string) {
 	const (
 		version  = "2.46.0"

internal/rosa/hakurei.go

@@ -2,7 +2,19 @@ package rosa
 
 import "hakurei.app/internal/pkg"
 
-func (t Toolchain) newHakurei(suffix, script string) pkg.Artifact {
+func (t Toolchain) newHakurei(
+	suffix, script string,
+	withHostname bool,
+) pkg.Artifact {
+	hostname := `
+echo '# Building test helper (hostname).'
+go build -v -o /bin/hostname /usr/src/hostname/main.go
+echo
+`
+	if !withHostname {
+		hostname = ""
+	}
+
 	return t.New("hakurei"+suffix+"-"+hakureiVersion, 0, []pkg.Artifact{
 		t.Load(Go),
 
@@ -29,17 +41,12 @@ func (t Toolchain) newHakurei(suffix, script string) pkg.Artifact {
 		"CGO_ENABLED=1",
 		"GOCACHE=/tmp/gocache",
 		"CC=clang -O3 -Werror",
-	}, `
-echo '# Building test helper (hostname).'
-go build -v -o /bin/hostname /usr/src/hostname/main.go
-echo
-
-chmod -R +w /usr/src/hakurei
+	}, hostname+`
 cd /usr/src/hakurei
 
 HAKUREI_VERSION='v`+hakureiVersion+`'
 `+script, pkg.Path(AbsUsrSrc.Append("hakurei"), true, t.NewPatchedSource(
-		"hakurei", hakureiVersion, hakureiSource, true, hakureiPatches...,
+		"hakurei", hakureiVersion, hakureiSource, false, hakureiPatches...,
 	)), pkg.Path(AbsUsrSrc.Append("hostname", "main.go"), false, pkg.NewFile(
 		"hostname.go",
 		[]byte(`
@@ -69,10 +76,11 @@ go build -trimpath -v -o /work/system/libexec/hakurei -ldflags="-s -w
 	-buildid=
 	-linkmode external
 	-extldflags=-static
-	-X hakurei.app/internal/info.buildVersion="$HAKUREI_VERSION"
+	-X hakurei.app/internal/info.buildVersion=${HAKUREI_VERSION}
 	-X hakurei.app/internal/info.hakureiPath=/system/bin/hakurei
 	-X hakurei.app/internal/info.hsuPath=/system/bin/hsu
-	-X main.hakureiPath=/system/bin/hakurei" ./...
+	-X main.hakureiPath=/system/bin/hakurei
+" ./...
 echo
 
 echo '# Testing hakurei.'
@@ -84,7 +92,7 @@ mkdir -p /work/system/bin/
 	hakurei \
 	sharefs \
 	../../bin/)
-`), hakureiVersion
+`, true), hakureiVersion
 		},
 
 		Name:        "hakurei",
@@ -98,7 +106,7 @@ mkdir -p /work/system/bin/
 			return t.newHakurei("-dist", `
 export HAKUREI_VERSION
 DESTDIR=/work /usr/src/hakurei/dist/release.sh
-`), hakureiVersion
+`, true), hakureiVersion
 		},
 
 		Name:        "hakurei-dist",

internal/rosa/hakurei_release.go

@@ -4,48 +4,15 @@ package rosa
 
 import "hakurei.app/internal/pkg"
 
-const hakureiVersion = "0.3.5"
+const hakureiVersion = "0.3.6"
 
 // hakureiSource is the source code of a hakurei release.
 var hakureiSource = pkg.NewHTTPGetTar(
 	nil, "https://git.gensokyo.uk/security/hakurei/archive/"+
 		"v"+hakureiVersion+".tar.gz",
-	mustDecode("6Tn38NLezRD2d3aGdFg5qFfqn8_KvC6HwMKwJMPvaHmVw8xRgxn8B0PObswl2mOk"),
+	mustDecode("Yul9J2yV0x453lQP9KUnG_wEJo_DbKMNM7xHJGt4rITCSeX9VRK2J4kzAxcv_0-b"),
 	pkg.TarGzip,
 )
 
 // hakureiPatches are patches applied against a hakurei release.
-var hakureiPatches = [][2]string{
-	{"createTemp-error-injection", `diff --git a/container/dispatcher_test.go b/container/dispatcher_test.go
-index 5de37fc..fe0c4db 100644
---- a/container/dispatcher_test.go
-+++ b/container/dispatcher_test.go
-@@ -238,8 +238,11 @@ func sliceAddr[S any](s []S) *[]S { return &s }
-
- func newCheckedFile(t *testing.T, name, wantData string, closeErr error) osFile {
-        f := &checkedOsFile{t: t, name: name, want: wantData, closeErr: closeErr}
--       // check happens in Close, and cleanup is not guaranteed to run, so relying on it for sloppy implementations will cause sporadic test results
--       f.cleanup = runtime.AddCleanup(f, func(name string) { f.t.Fatalf("checkedOsFile %s became unreachable without a call to Close", name) }, f.name)
-+       // check happens in Close, and cleanup is not guaranteed to run, so relying
-+       // on it for sloppy implementations will cause sporadic test results
-+       f.cleanup = runtime.AddCleanup(f, func(name string) {
-+               panic("checkedOsFile " + name + " became unreachable without a call to Close")
-+       }, name)
-        return f
- }
-
-diff --git a/container/initplace_test.go b/container/initplace_test.go
-index afeddbe..1c2f20b 100644
---- a/container/initplace_test.go
-+++ b/container/initplace_test.go
-@@ -21,7 +21,7 @@ func TestTmpfileOp(t *testing.T) {
-                        Path: samplePath,
-                        Data: sampleData,
-                }, nil, nil, []stub.Call{
--                       call("createTemp", stub.ExpectArgs{"/", "tmp.*"}, newCheckedFile(t, "tmp.32768", sampleDataString, nil), stub.UniqueError(5)),
-+                       call("createTemp", stub.ExpectArgs{"/", "tmp.*"}, (*checkedOsFile)(nil), stub.UniqueError(5)),
-                }, stub.UniqueError(5)},
-
-                {"Write", &Params{ParentPerm: 0700}, &TmpfileOp{
-`},
-}
+var hakureiPatches [][2]string

internal/rosa/images.go

@@ -2,10 +2,32 @@ package rosa
 
 import "hakurei.app/internal/pkg"
 
+func init() {
+	artifactsM[EarlyInit] = Metadata{
+		Name:        "earlyinit",
+		Description: "Rosa OS initramfs init program",
+
+		f: func(t Toolchain) (pkg.Artifact, string) {
+			return t.newHakurei("-early-init", `
+mkdir -p /work/system/libexec/hakurei/
+
+echo '# Building earlyinit.'
+go build -trimpath -v -o /work/system/libexec/hakurei -ldflags="-s -w
+	-buildid=
+	-linkmode external
+	-extldflags=-static
+	-X hakurei.app/internal/info.buildVersion=${HAKUREI_VERSION}
+" ./cmd/earlyinit
+echo
+`, false), Unversioned
+		},
+	}
+}
+
 func (t Toolchain) newImageInitramfs() (pkg.Artifact, string) {
 	return t.New("initramfs", TNoToolchain, []pkg.Artifact{
 		t.Load(Zstd),
-		t.Load(Hakurei),
+		t.Load(EarlyInit),
 		t.Load(GenInitCPIO),
 	}, nil, nil, `
 gen_init_cpio -t 4294967295 -c /usr/src/initramfs | zstd > /work/initramfs.zst

internal/rosa/kernel.go

@@ -82,6 +82,11 @@ install -Dm0500 \
 echo "Installing linux $1..."
 cp -av "$2" "$4"
 cp -av "$3" "$4"
+`))),
+			pkg.Path(AbsUsrSrc.Append(
+				".depmod",
+			), false, pkg.NewFile("depmod", []byte(`#!/bin/sh
+exec /system/sbin/depmod -m /lib/modules "$@"
 `))),
 		},
 
@@ -1210,6 +1215,11 @@ cgit 1.2.3-korg
 			"all",
 		},
 		Install: `
+# kernel is not aware of kmod moduledir
+install -Dm0500 \
+	/usr/src/.depmod \
+	/sbin/depmod
+
 make \
 	"-j$(nproc)" \
 	-f /usr/src/kernel/Makefile \
@@ -1217,9 +1227,10 @@ make \
 	LLVM=1 \
 	INSTALL_PATH=/work \
 	install \
-	INSTALL_MOD_PATH=/work \
+	INSTALL_MOD_PATH=/work/system \
+	DEPMOD=/sbin/depmod \
 	modules_install
-rm -v /work/lib/modules/` + kernelVersion + `/build
+rm -v /work/system/lib/modules/` + kernelVersion + `/build
 `,
 	},
 		Flex,
@@ -1272,3 +1283,55 @@ func init() {
 		Description: "a program in the kernel source tree for creating initramfs archive",
 	}
 }
+
+func (t Toolchain) newFirmware() (pkg.Artifact, string) {
+	const (
+		version  = "20260221"
+		checksum = "vTENPW5rZ6yLVq7YKDLHkCVgKXvwUWigEx7T4LcxoKeBVYIyf1_sEExeV4mo-e46"
+	)
+	return t.NewPackage("firmware", version, pkg.NewHTTPGetTar(
+		nil, "https://gitlab.com/kernel-firmware/linux-firmware/-/"+
+			"archive/"+version+"/linux-firmware-"+version+".tar.bz2",
+		mustDecode(checksum),
+		pkg.TarBzip2,
+	), &PackageAttr{
+		// dedup creates temporary file
+		Writable: true,
+		// does not use configure
+		EnterSource: true,
+
+		Env: []string{
+			"HOME=/proc/nonexistent",
+		},
+	}, &MakeHelper{
+		OmitDefaults:  true,
+		SkipConfigure: true,
+		InPlace:       true,
+
+		Make: []string{
+			"DESTDIR=/work/system",
+			"install-zst",
+		},
+		SkipCheck: true, // requires pre-commit
+		Install:   `make "-j$(nproc)" DESTDIR=/work/system dedup`,
+	},
+		Perl,
+		Parallel,
+		Nettle,
+		Rdfind,
+		Zstd,
+		Findutils,
+		Coreutils,
+	), version
+}
+func init() {
+	artifactsM[Firmware] = Metadata{
+		f: Toolchain.newFirmware,
+
+		Name:        "firmware",
+		Description: "firmware blobs for use with the Linux kernel",
+		Website:     "https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git/",
+
+		ID: 141464,
+	}
+}

internal/rosa/kernel_arm64.config

@@ -1,16 +1,16 @@
 #
 # Automatically generated file; DO NOT EDIT.
-# Linux/arm64 6.12.73 Kernel Configuration
+# Linux/arm64 6.12.76 Kernel Configuration
 #
-CONFIG_CC_VERSION_TEXT="clang version 21.1.8"
+CONFIG_CC_VERSION_TEXT="clang version 22.1.0"
 CONFIG_GCC_VERSION=0
 CONFIG_CC_IS_CLANG=y
-CONFIG_CLANG_VERSION=210108
+CONFIG_CLANG_VERSION=220100
 CONFIG_AS_IS_LLVM=y
-CONFIG_AS_VERSION=210108
+CONFIG_AS_VERSION=220100
 CONFIG_LD_VERSION=0
 CONFIG_LD_IS_LLD=y
-CONFIG_LLD_VERSION=210108
+CONFIG_LLD_VERSION=220100
 CONFIG_RUSTC_VERSION=0
 CONFIG_RUSTC_LLVM_VERSION=0
 CONFIG_CC_HAS_ASM_GOTO_OUTPUT=y
@@ -4984,7 +4984,7 @@ CONFIG_SERIAL_TEGRA_TCU=m
 CONFIG_SERIAL_MAX3100=m
 CONFIG_SERIAL_MAX310X=m
 CONFIG_SERIAL_IMX=m
-CONFIG_SERIAL_IMX_CONSOLE=m
+# CONFIG_SERIAL_IMX_CONSOLE is not set
 CONFIG_SERIAL_IMX_EARLYCON=y
 CONFIG_SERIAL_UARTLITE=m
 CONFIG_SERIAL_UARTLITE_NR_UARTS=1
@@ -5772,6 +5772,7 @@ CONFIG_GPIO_MADERA=m
 CONFIG_GPIO_MAX77650=m
 CONFIG_GPIO_PMIC_EIC_SPRD=m
 CONFIG_GPIO_SL28CPLD=m
+CONFIG_GPIO_TN48M_CPLD=m
 CONFIG_GPIO_TPS65086=m
 CONFIG_GPIO_TPS65218=m
 CONFIG_GPIO_TPS65219=m
@@ -6471,6 +6472,7 @@ CONFIG_MFD_MAX5970=m
 # CONFIG_MFD_CS47L85 is not set
 # CONFIG_MFD_CS47L90 is not set
 # CONFIG_MFD_CS47L92 is not set
+CONFIG_MFD_TN48M_CPLD=m
 # CONFIG_MFD_DA9052_SPI is not set
 CONFIG_MFD_DA9062=m
 CONFIG_MFD_DA9063=m
@@ -12532,6 +12534,7 @@ CONFIG_RESET_SUNXI=y
 CONFIG_RESET_TI_SCI=m
 CONFIG_RESET_TI_SYSCON=m
 CONFIG_RESET_TI_TPS380X=m
+CONFIG_RESET_TN48M_CPLD=m
 CONFIG_RESET_UNIPHIER=m
 CONFIG_RESET_UNIPHIER_GLUE=m
 CONFIG_RESET_ZYNQMP=y
@@ -14022,7 +14025,6 @@ CONFIG_LOCK_DEBUGGING_SUPPORT=y
 
 # CONFIG_DEBUG_IRQFLAGS is not set
 CONFIG_STACKTRACE=y
-# CONFIG_WARN_ALL_UNSEEDED_RANDOM is not set
 # CONFIG_DEBUG_KOBJECT is not set
 
 #
@@ -14057,7 +14059,7 @@ CONFIG_USER_STACKTRACE_SUPPORT=y
 CONFIG_NOP_TRACER=y
 CONFIG_HAVE_FUNCTION_TRACER=y
 CONFIG_HAVE_FUNCTION_GRAPH_TRACER=y
-CONFIG_HAVE_FUNCTION_GRAPH_RETVAL=y
+CONFIG_HAVE_FUNCTION_GRAPH_FREGS=y
 CONFIG_HAVE_DYNAMIC_FTRACE=y
 CONFIG_HAVE_DYNAMIC_FTRACE_WITH_ARGS=y
 CONFIG_HAVE_FTRACE_MCOUNT_RECORD=y

internal/rosa/kmod.go

@@ -14,6 +14,7 @@ func (t Toolchain) newKmod() (pkg.Artifact, string) {
 		pkg.TarGzip,
 	), nil, &MesonHelper{
 		Setup: [][2]string{
+			{"Dmoduledir", "/system/lib/modules"},
 			{"Dsysconfdir", "/system/etc"},
 			{"Dbashcompletiondir", "no"},
 			{"Dfishcompletiondir", "no"},

internal/rosa/libxslt.go

@@ -23,6 +23,7 @@ func (t Toolchain) newLibxslt() (pkg.Artifact, string) {
 		SkipCheck: true,
 	},
 		XZ,
+		Zlib,
 		Python,
 		PkgConfig,
 

internal/rosa/llvm.go

@@ -125,6 +125,8 @@ func (t Toolchain) newLLVMVariant(variant string, attr *llvmAttr) pkg.Artifact {
 
 			[2]string{"LLVM_INSTALL_BINUTILS_SYMLINKS", "ON"},
 			[2]string{"LLVM_INSTALL_CCTOOLS_SYMLINKS", "ON"},
+
+			[2]string{"LLVM_LIT_ARGS", "'--verbose'"},
 		)
 	}
 
@@ -187,6 +189,7 @@ ln -s ld.lld /work/system/bin/ld
 		Append: cmakeAppend,
 		Script: script + attr.script,
 	},
+		Zlib,
 		Libffi,
 		Python,
 		Perl,

internal/rosa/meson.go

@@ -13,6 +13,7 @@ func (t Toolchain) newMeson() (pkg.Artifact, string) {
 		checksum = "w895BXF_icncnXatT_OLCFe2PYEtg4KrKooMgUYdN-nQVvbFX3PvYWHGEpogsHtd"
 	)
 	return t.New("meson-"+version, 0, []pkg.Artifact{
+		t.Load(Zlib),
 		t.Load(Python),
 		t.Load(Setuptools),
 	}, nil, nil, `
@@ -66,6 +67,7 @@ func (*MesonHelper) name(name, version string) string {
 // extra returns hardcoded meson runtime dependencies.
 func (*MesonHelper) extra(int) []PArtifact {
 	return []PArtifact{
+		Zlib,
 		Python,
 		Meson,
 		Ninja,

internal/rosa/nettle.go

@@ -0,0 +1,31 @@
+package rosa
+
+import "hakurei.app/internal/pkg"
+
+func (t Toolchain) newNettle() (pkg.Artifact, string) {
+	const (
+		version  = "4.0"
+		checksum = "6agC-vHzzoqAlaX3K9tX8yHgrm03HLqPZzVzq8jh_ePbuPMIvpxereu_uRJFmQK7"
+	)
+	return t.NewPackage("nettle", version, pkg.NewHTTPGetTar(
+		nil, "https://ftpmirror.gnu.org/gnu/nettle/nettle-"+version+".tar.gz",
+		mustDecode(checksum),
+		pkg.TarGzip,
+	), nil, (*MakeHelper)(nil),
+		M4,
+		Diffutils,
+
+		GMP,
+	), version
+}
+func init() {
+	artifactsM[Nettle] = Metadata{
+		f: Toolchain.newNettle,
+
+		Name:        "nettle",
+		Description: "a low-level cryptographic library",
+		Website:     "https://www.lysator.liu.se/~nisse/nettle/",
+
+		ID: 2073,
+	}
+}

internal/rosa/nss.go

@@ -84,7 +84,7 @@ func init() {
 	artifactsM[buildcatrust] = newViaPip(
 		"buildcatrust",
 		"transform certificate stores between formats",
-		version, "none", "any",
+		version, "py3", "none", "any",
 		"k_FGzkRCLjbTWBkuBLzQJ1S8FPAz19neJZlMHm0t10F2Y0hElmvVwdSBRc03Rjo1",
 		"https://github.com/nix-community/buildcatrust/"+
 			"releases/download/v"+version+"/",
@@ -93,6 +93,7 @@ func init() {
 
 func (t Toolchain) newNSSCACert() (pkg.Artifact, string) {
 	return t.New("nss-cacert", 0, []pkg.Artifact{
+		t.Load(Zlib),
 		t.Load(Bash),
 		t.Load(Python),
 

internal/rosa/python.go

@@ -75,10 +75,10 @@ func init() {
 
 // newViaPip is a helper for installing python dependencies via pip.
 func newViaPip(
-	name, description, version, abi, platform, checksum, prefix string,
+	name, description, version, interpreter, abi, platform, checksum, prefix string,
 	extra ...PArtifact,
 ) Metadata {
-	wname := name + "-" + version + "-py3-" + abi + "-" + platform + ".whl"
+	wname := name + "-" + version + "-" + interpreter + "-" + abi + "-" + platform + ".whl"
 	return Metadata{
 		f: func(t Toolchain) (pkg.Artifact, string) {
 			extraRes := make([]pkg.Artifact, len(extra))
@@ -87,6 +87,7 @@ func newViaPip(
 			}
 
 			return t.New(name+"-"+version, 0, slices.Concat([]pkg.Artifact{
+				t.Load(Zlib),
 				t.Load(Python),
 			}, extraRes), nil, nil, `
 pip3 install \
@@ -112,6 +113,7 @@ func (t Toolchain) newSetuptools() (pkg.Artifact, string) {
 		checksum = "K9f8Yi7Gg95zjmQsE1LLw9UBb8NglI6EY6pQpdD6DM0Pmc_Td5w2qs1SMngTI6Jp"
 	)
 	return t.New("setuptools-"+version, 0, []pkg.Artifact{
+		t.Load(Zlib),
 		t.Load(Python),
 	}, nil, nil, `
 pip3 install \
@@ -142,7 +144,7 @@ func init() {
 	artifactsM[PythonPygments] = newViaPip(
 		"pygments",
 		" a syntax highlighting package written in Python",
-		"2.19.2", "none", "any",
+		"2.19.2", "py3", "none", "any",
 		"ak_lwTalmSr7W4Mjy2XBZPG9I6a0gwSy2pS87N8x4QEuZYif0ie9z0OcfRfi9msd",
 		"https://files.pythonhosted.org/packages/"+
 			"c7/21/705964c7812476f378728bdf590ca4b771ec72385c533964653c68e86bdc/",
@@ -151,7 +153,7 @@ func init() {
 	artifactsM[PythonPluggy] = newViaPip(
 		"pluggy",
 		"the core framework used by the pytest, tox, and devpi projects",
-		"1.6.0", "none", "any",
+		"1.6.0", "py3", "none", "any",
 		"2HWYBaEwM66-y1hSUcWI1MyE7dVVuNNRW24XD6iJBey4YaUdAK8WeXdtFMQGC-4J",
 		"https://files.pythonhosted.org/packages/"+
 			"54/20/4d324d65cc6d9205fabedc306948156824eb9f0ee1633355a8f7ec5c66bf/",
@@ -160,7 +162,7 @@ func init() {
 	artifactsM[PythonPackaging] = newViaPip(
 		"packaging",
 		"reusable core utilities for various Python Packaging interoperability specifications",
-		"26.0", "none", "any",
+		"26.0", "py3", "none", "any",
 		"iVVXcqdwHDskPKoCFUlh2x8J0Gyq-bhO4ns9DvUJ7oJjeOegRYtSIvLV33Bki-pP",
 		"https://files.pythonhosted.org/packages/"+
 			"b7/b9/c538f279a4e237a006a2c98387d081e9eb060d203d8ed34467cc0f0b9b53/",
@@ -169,15 +171,16 @@ func init() {
 	artifactsM[PythonIniConfig] = newViaPip(
 		"iniconfig",
 		"a small and simple INI-file parser module",
-		"2.3.0", "none", "any",
+		"2.3.0", "py3", "none", "any",
 		"SDgs4S5bXi77aVOeKTPv2TUrS3M9rduiK4DpU0hCmDsSBWqnZcWInq9lsx6INxut",
 		"https://files.pythonhosted.org/packages/"+
 			"cb/b1/3846dd7f199d53cb17f49cba7e651e9ce294d8497c8c150530ed11865bb8/",
 	)
+
 	artifactsM[PythonPyTest] = newViaPip(
 		"pytest",
 		"the pytest framework",
-		"9.0.2", "none", "any",
+		"9.0.2", "py3", "none", "any",
 		"IM2wDbLke1EtZhF92zvAjUl_Hms1uKDtM7U8Dt4acOaChMnDg1pW7ib8U0wYGDLH",
 		"https://files.pythonhosted.org/packages/"+
 			"3b/ab/b3226f0bd7cdcf710fbede2b3548584366da3b19b5021e74f5bde2a8fa3f/",
@@ -186,4 +189,109 @@ func init() {
 		PythonPluggy,
 		PythonPygments,
 	)
+
+	artifactsM[PythonCfgv] = newViaPip(
+		"cfgv",
+		"validate configuration and produce human readable error messages",
+		"3.5.0", "py2.py3", "none", "any",
+		"yFKTyVRlmnLKAxvvge15kAd_GOP1Xh3fZ0NFImO5pBdD5e0zj3GRmA6Q1HdtLTYO",
+		"https://files.pythonhosted.org/packages/"+
+			"db/3c/33bac158f8ab7f89b2e59426d5fe2e4f63f7ed25df84c036890172b412b5/",
+	)
+
+	artifactsM[PythonIdentify] = newViaPip(
+		"identify",
+		"file identification library for Python",
+		"2.6.17", "py2.py3", "none", "any",
+		"9RxK3igO-Pxxof5AuCAGiF_L1SWi4SpuSF1fWNXCzE2D4oTRSob-9VpFMLlybrSv",
+		"https://files.pythonhosted.org/packages/"+
+			"40/66/71c1227dff78aaeb942fed29dd5651f2aec166cc7c9aeea3e8b26a539b7d/",
+	)
+
+	artifactsM[PythonNodeenv] = newViaPip(
+		"nodeenv",
+		"a tool to create isolated node.js environments",
+		"1.10.0", "py2.py3", "none", "any",
+		"ihUb4-WQXYIhYOOKSsXlKIzjzQieOYl6ojro9H-0DFzGheaRTtuyZgsCmriq58sq",
+		"https://files.pythonhosted.org/packages/"+
+			"88/b2/d0896bdcdc8d28a7fc5717c305f1a861c26e18c05047949fb371034d98bd/",
+	)
+
+	artifactsM[PythonPyYAML] = newViaPip(
+		"pyyaml",
+		"a complete YAML 1.1 parser",
+		"6.0.3", "cp314", "cp314", "musllinux_1_2_x86_64",
+		"4_jhCFpUNtyrFp2HOMqUisR005u90MHId53eS7rkUbcGXkoaJ7JRsY21dREHEfGN",
+		"https://files.pythonhosted.org/packages/"+
+			"d7/ce/af88a49043cd2e265be63d083fc75b27b6ed062f5f9fd6cdc223ad62f03e/",
+	)
+
+	artifactsM[PythonDistlib] = newViaPip(
+		"distlib",
+		"used as the basis for third-party packaging tools",
+		"0.4.0", "py2.py3", "none", "any",
+		"lGLLfYVhUhXOTw_84zULaH2K8n6pk1OOVXmJfGavev7N42msbtHoq-XY5D_xULI_",
+		"https://files.pythonhosted.org/packages/"+
+			"33/6b/e0547afaf41bf2c42e52430072fa5658766e3d65bd4b03a563d1b6336f57/",
+	)
+
+	artifactsM[PythonFilelock] = newViaPip(
+		"filelock",
+		"a platform-independent file locking library for Python",
+		"3.25.0", "py3", "none", "any",
+		"0gSQIYNUEjOs1JBxXjGwfLnwFPFINwqyU_Zqgj7fT_EGafv_HaD5h3Xv2Rq_qQ44",
+		"https://files.pythonhosted.org/packages/"+
+			"f9/0b/de6f54d4a8bedfe8645c41497f3c18d749f0bd3218170c667bf4b81d0cdd/",
+	)
+
+	artifactsM[PythonPlatformdirs] = newViaPip(
+		"platformdirs",
+		"a Python package for determining platform-specific directories",
+		"4.9.4", "py3", "none", "any",
+		"JGNpMCX2JMn-7c9bk3QzOSNDgJRR_5lH-jIqfy0zXMZppRCdLsTNbdp4V7QFwxOI",
+		"https://files.pythonhosted.org/packages/"+
+			"63/d7/97f7e3a6abb67d8080dd406fd4df842c2be0efaf712d1c899c32a075027c/",
+	)
+
+	artifactsM[PythonDiscovery] = newViaPip(
+		"python_discovery",
+		"looks for a python installation",
+		"1.1.1", "py3", "none", "any",
+		"Jk_qGMfZYm0fdNOSvMdVQZuQbJlqu3NWRm7T2fRtiBXmHLQyOdJE3ypI_it1OJR0",
+		"https://files.pythonhosted.org/packages/"+
+			"75/0f/2bf7e3b5a4a65f623cb820feb5793e243fad58ae561015ee15a6152f67a2/",
+		PythonFilelock,
+		PythonPlatformdirs,
+	)
+
+	artifactsM[PythonVirtualenv] = newViaPip(
+		"virtualenv",
+		"a tool for creating isolated virtual python environments",
+		"21.1.0", "py3", "none", "any",
+		"SLvdr3gJZ7GTS-kiRyq2RvJdrQ8SZYC1pglbViWCMLCuAIcbLNjVEUJZ4hDtKUxm",
+		"https://files.pythonhosted.org/packages/"+
+			"78/55/896b06bf93a49bec0f4ae2a6f1ed12bd05c8860744ac3a70eda041064e4d/",
+		PythonDistlib,
+		PythonFilelock,
+		PythonPlatformdirs,
+		PythonDiscovery,
+	)
+
+	artifactsM[PythonPreCommit] = newViaPip(
+		"pre_commit",
+		"a framework for managing and maintaining multi-language pre-commit hooks",
+		"4.5.1", "py2.py3", "none", "any",
+		"9G2Hv5JpvXFZVfw4pv_KAsmHD6bvot9Z0YBDmW6JeJizqTA4xEQCKel-pCERqQFK",
+		"https://files.pythonhosted.org/packages/"+
+			"5d/19/fd3ef348460c80af7bb4669ea7926651d1f95c23ff2df18b9d24bab4f3fa/",
+		PythonCfgv,
+		PythonIdentify,
+		PythonNodeenv,
+		PythonPyYAML,
+		PythonDistlib,
+		PythonFilelock,
+		PythonPlatformdirs,
+		PythonDiscovery,
+		PythonVirtualenv,
+	)
 }

internal/rosa/rdfind.go

@@ -0,0 +1,33 @@
+package rosa
+
+import "hakurei.app/internal/pkg"
+
+func (t Toolchain) newRdfind() (pkg.Artifact, string) {
+	const (
+		version  = "1.8.0"
+		checksum = "PoaeJ2WIG6yyfe5VAYZlOdAQiR3mb3WhAUMj2ziTCx_IIEal4640HMJUb4SzU9U3"
+	)
+	return t.NewPackage("rdfind", version, pkg.NewHTTPGetTar(
+		nil, "https://rdfind.pauldreik.se/rdfind-"+version+".tar.gz",
+		mustDecode(checksum),
+		pkg.TarGzip,
+	), nil, &MakeHelper{
+		// test suite hard codes /bin/echo
+		ScriptCheckEarly: `
+ln -s ../system/bin/toybox /bin/echo
+`,
+	},
+		Nettle,
+	), version
+}
+func init() {
+	artifactsM[Rdfind] = Metadata{
+		f: Toolchain.newRdfind,
+
+		Name:        "rdfind",
+		Description: "a program that finds duplicate files",
+		Website:     "https://rdfind.pauldreik.se/",
+
+		ID: 231641,
+	}
+}

internal/rosa/stage0.go

@@ -15,6 +15,7 @@ func (t Toolchain) newStage0() (pkg.Artifact, string) {
 		runtimes,
 		clang,
 
+		t.Load(Zlib),
 		t.Load(Bzip2),
 
 		t.Load(Patch),

internal/rosa/tamago.go

@@ -44,5 +44,7 @@ func init() {
 		Name:        "tamago",
 		Description: "a Go toolchain extended with support for bare metal execution",
 		Website:     "https://github.com/usbarmory/tamago-go",
+
+		ID: 388872,
 	}
 }

internal/rosa/zlib.go

@@ -4,22 +4,28 @@ import "hakurei.app/internal/pkg"
 
 func (t Toolchain) newZlib() (pkg.Artifact, string) {
 	const (
-		version  = "1.3.1"
-		checksum = "E-eIpNzE8oJ5DsqH4UuA_0GDKuQF5csqI8ooDx2w7Vx-woJ2mb-YtSbEyIMN44mH"
+		version  = "1.3.2"
+		checksum = "KHZrePe42vL2XvOUE3KlJkp1UgWhWkl0jjT_BOvFhuM4GzieEH9S7CioepOFVGYB"
 	)
 	return t.NewPackage("zlib", version, pkg.NewHTTPGetTar(
 		nil, "https://www.zlib.net/fossils/zlib-"+version+".tar.gz",
 		mustDecode(checksum),
 		pkg.TarGzip,
-	), &PackageAttr{
-		Env: []string{
-			"CC=clang -fPIC",
-		},
-	}, &MakeHelper{
-		OmitDefaults: true,
+	), nil, &CMakeHelper{
+		Cache: [][2]string{
+			{"CMAKE_BUILD_TYPE", "Release"},
 
-		Host:  `""`,
-		Build: `""`,
+			{"ZLIB_BUILD_TESTING", "OFF"},
+			{"ZLIB_BUILD_SHARED", "ON"},
+			{"ZLIB_BUILD_STATIC", "ON"},
+			{"ZLIB_BUILD_MINIZIP", "OFF"},
+
+			{"ZLIB_INSTALL", "ON"},
+			{"ZLIB_PREFIX", "OFF"},
+		},
+
+		// ninja dependency loop
+		Make: true,
 	}), version
 }
 func init() {

internal/rosa/zstd.go

@@ -16,7 +16,6 @@ func (t Toolchain) newZstd() (pkg.Artifact, string) {
 		Append: []string{"build", "cmake"},
 		Cache: [][2]string{
 			{"CMAKE_BUILD_TYPE", "Release"},
-			{"CMAKE_INSTALL_LIBDIR", "lib"},
 		},
 	}), version
 }

nixos.nix

@@ -139,6 +139,8 @@ in
                     inherit (app) identity groups enablements;
                     inherit (dbusConfig) session_bus system_bus;
                     direct_wayland = app.insecureWayland;
+                    sched_policy = app.schedPolicy;
+                    sched_priority = app.schedPriority;
 
                     container = {
                       inherit (app)

options.nix

@@ -98,6 +98,7 @@ in
               ints
               str
               bool
+              enum
               package
               anything
               submodule
@@ -237,6 +238,29 @@ in
               };
               hostAbstract = mkEnableOption "share abstract unix socket scope";
 
+              schedPolicy = mkOption {
+                type = nullOr (enum [
+                  "fifo"
+                  "rr"
+                  "batch"
+                  "idle"
+                  "deadline"
+                  "ext"
+                ]);
+                default = null;
+                description = ''
+                  Scheduling policy to set for the container.
+                  The zero value retains the current scheduling policy.
+                '';
+              };
+              schedPriority = mkOption {
+                type = nullOr (ints.between 1 99);
+                default = null;
+                description = ''
+                  Scheduling priority to set for the container.
+                '';
+              };
+
               nix = mkEnableOption "nix daemon access";
               mapRealUid = mkEnableOption "mapping to priv-user uid";
               device = mkEnableOption "access to all devices";

package.nix

@@ -1,7 +1,7 @@
 {
   lib,
   stdenv,
-  buildGoModule,
+  buildGo126Module,
   makeBinaryWrapper,
   xdg-dbus-proxy,
   pkg-config,
@@ -17,7 +17,7 @@
   fuse3,
 
   # for passthru.buildInputs
-  go,
+  go_1_26,
   clang,
 
   # for check
@@ -28,7 +28,7 @@
   withStatic ? stdenv.hostPlatform.isStatic,
 }:
 
-buildGoModule rec {
+buildGo126Module rec {
   pname = "hakurei";
   version = "0.3.6";
 
@@ -51,7 +51,7 @@ buildGoModule rec {
     ];
 
     nativeBuildInputs = [
-      go
+      go_1_26
       pkg-config
       wayland-scanner
     ];
@@ -125,16 +125,20 @@ buildGoModule rec {
         --inherit-argv0 --prefix PATH : ${lib.makeBinPath appPackages}
     '';
 
-  passthru.targetPkgs = [
-    go
-    clang
-    xorg.xorgproto
-    util-linux
+  passthru = {
+    go = go_1_26;
 
-    # for go generate
-    wayland-protocols
-    wayland-scanner
-  ]
-  ++ buildInputs
-  ++ nativeBuildInputs;
+    targetPkgs = [
+      go_1_26
+      clang
+      xorg.xorgproto
+      util-linux
+
+      # for go generate
+      wayland-protocols
+      wayland-scanner
+    ]
+    ++ buildInputs
+    ++ nativeBuildInputs;
+  };
 }

test/configuration.nix

@@ -28,6 +28,15 @@
   # Automatically login on tty1 as a normal user:
   services.getty.autologinUser = "alice";
 
+  security.pam.loginLimits = [
+    {
+      domain = "@users";
+      item = "rtprio";
+      type = "-";
+      value = 1;
+    }
+  ];
+
   environment = {
     systemPackages = with pkgs; [
       # For D-Bus tests:

test/default.nix

@@ -34,7 +34,7 @@ testers.nixosTest {
         (writeShellScriptBin "hakurei-test" ''
           # Assert hst CGO_ENABLED=0: ${
             with pkgs;
-            runCommand "hakurei-hst-cgo" { nativeBuildInputs = [ go ]; } ''
+            runCommand "hakurei-hst-cgo" { nativeBuildInputs = [ self.packages.${system}.hakurei.go ]; } ''
               cp -r ${options.environment.hakurei.package.default.src} "$out"
               chmod -R +w "$out"
               cp ${writeText "hst_cgo_test.go" ''package hakurei_test;import("testing";"hakurei.app/hst");func TestTemplate(t *testing.T){hst.Template()}''} "$out/hst_cgo_test.go"

test/interactive/configuration.nix

@@ -23,6 +23,14 @@
   security = {
     sudo.wheelNeedsPassword = false;
     rtkit.enable = true;
+    pam.loginLimits = [
+      {
+        domain = "@users";
+        item = "rtprio";
+        type = "-";
+        value = 1;
+      }
+    ];
   };
 
   services = {

test/test.py

@@ -206,6 +206,17 @@ machine.wait_until_fails("pgrep foot", timeout=5)
 machine.wait_for_file("/tmp/shim-cont-unexpected-pid")
 print(machine.succeed('grep "shim: got SIGCONT from unexpected process$" /tmp/shim-cont-unexpected-pid'))
 
+# Check setscheduler:
+sched_unset = int(machine.succeed("sudo -u alice -i hakurei -v run cat /proc/self/sched | grep '^policy' | tr -d ' ' | cut -d ':' -f 2"))
+if sched_unset != 0:
+    raise Exception(f"unexpected unset policy: {sched_unset}")
+sched_idle = int(machine.succeed("sudo -u alice -i hakurei -v run --policy=idle cat /proc/self/sched | grep '^policy' | tr -d ' ' | cut -d ':' -f 2"))
+if sched_idle != 5:
+    raise Exception(f"unexpected idle policy: {sched_idle}")
+sched_rr = int(machine.succeed("sudo -u alice -i hakurei -v run --policy=rr cat /proc/self/sched | grep '^policy' | tr -d ' ' | cut -d ':' -f 2"))
+if sched_rr != 2:
+    raise Exception(f"unexpected round-robin policy: {sched_idle}")
+
 # Start app (foot) with Wayland enablement:
 swaymsg("exec ne-foot")
 wait_for_window(f"u0_a{hakurei_identity(0)}@machine")