@synopackageland/cli 0.3.10 → 0.4.1

package/templates/native-go/backend/internal/helper/control.go

@@ -0,0 +1,424 @@
+package helper
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"io"
+	"io/fs"
+	"net"
+	"os"
+	"path/filepath"
+	"strings"
+	"syscall"
+	"time"
+)
+
+// Process identity rejects reboot/PID reuse and executable replacement.
+// Linux Stop authenticates the owning supervisor socket; only that parent signals its child.
+type Process struct {
+	Inventory Inventory `json:"inventory"`
+	Mode      string    `json:"mode,omitempty"` // must be supervised; old unsupervised records fail closed
+	PID       int       `json:"pid"`
+	Boot      string    `json:"boot"`
+	Start     string    `json:"start"`
+	Device    uint64    `json:"device"`
+	Inode     uint64    `json:"inode"`
+}
+
+type Processes interface {
+	Stop(Process) error
+	Start(Inventory) error
+}
+type Controller struct {
+	Client
+	Processes Processes
+	Observe   Observation
+	Boot      func() (string, error)
+}
+
+func waitHealthy(ctx context.Context, probe func(context.Context) bool) error {
+	for ctx.Err() == nil {
+		if probe(ctx) {
+			return nil
+		}
+		timer := time.NewTimer(20 * time.Millisecond)
+		select {
+		case <-ctx.Done():
+			timer.Stop()
+			return ctx.Err()
+		case <-timer.C:
+		}
+	}
+	return ctx.Err()
+}
+
+func writeProtected(dir, name string, value any) error {
+	b, err := json.Marshal(value)
+	if err != nil {
+		return err
+	}
+	f, err := os.CreateTemp(dir, ".transition-*")
+	if err != nil {
+		return err
+	}
+	defer os.Remove(f.Name())
+	mode := fs.FileMode(0644)
+	if name == "observer-diagnostic.json" || name == "process.json" || strings.HasPrefix(name, "completion-") {
+		mode = 0600
+	}
+	if err := f.Chmod(mode); err != nil {
+		f.Close()
+		return err
+	}
+	if _, err := f.Write(b); err != nil {
+		f.Close()
+		return err
+	}
+	if err := f.Sync(); err != nil {
+		f.Close()
+		return err
+	}
+	if err := f.Close(); err != nil {
+		return err
+	}
+	if err := os.Rename(f.Name(), filepath.Join(dir, name)); err != nil {
+		return err
+	}
+	d, err := os.Open(dir)
+	if err != nil {
+		return err
+	}
+	defer d.Close()
+	return d.Sync()
+}
+
+func writeProcess(dir string, p Process) error { return writeProtected(dir, "process.json", p) }
+
+func readProcess(dir string, owner uint32) (Process, error) {
+	var p Process
+	if err := protected(dir, owner, true); err != nil {
+		return p, err
+	}
+	path := filepath.Join(dir, "process.json")
+	if err := protected(path, owner, false); err != nil {
+		return p, err
+	}
+	f, err := os.OpenFile(path, os.O_RDONLY|syscall.O_NOFOLLOW, 0)
+	if err != nil {
+		return p, err
+	}
+	defer f.Close()
+	d := json.NewDecoder(io.LimitReader(f, 4097))
+	d.DisallowUnknownFields()
+	if d.Decode(&p) != nil || d.Decode(new(any)) != io.EOF || !p.Inventory.Valid() || p.Mode != "supervised" || p.PID <= 1 || p.Boot == "" || p.Start == "" || p.Inode == 0 {
+		return p, errTrust
+	}
+	return p, nil
+}
+
+// Control is called only by the verified root executable from a fixed DSM
+// task. Package hooks and CGI handlers never call this function.
+func (c Controller) Control(ctx context.Context, action string) error {
+	if ctx.Err() != nil {
+		return ctx.Err()
+	}
+	if action != "enable" && action != "recover" && action != "teardown" && action != "disable" && action != "repair" {
+		return errors.New("unsupported lifecycle action")
+	}
+	if !c.Expected.Valid() || c.Processes == nil {
+		return errTrust
+	}
+	if err := protected(c.StateDir, c.Owner, true); err != nil {
+		return err
+	}
+	f, err := os.OpenFile(filepath.Join(c.StateDir, "control.lock"), os.O_CREATE|os.O_RDWR|syscall.O_NOFOLLOW, 0600)
+	if err != nil {
+		return err
+	}
+	defer f.Close()
+	if err := protected(f.Name(), c.Owner, false); err != nil {
+		return err
+	}
+	if err := syscall.Flock(int(f.Fd()), syscall.LOCK_EX|syscall.LOCK_NB); err != nil {
+		return errors.New("lifecycle busy")
+	}
+	defer syscall.Flock(int(f.Fd()), syscall.LOCK_UN)
+	r, err := readRecord(c.StateDir, c.Owner)
+	if err != nil && !errors.Is(err, fs.ErrNotExist) {
+		return err
+	}
+	if errors.Is(err, fs.ErrNotExist) {
+		r = Record{Inventory: c.Expected}
+	}
+	if action == "repair" {
+		if err != nil {
+			return err
+		}
+		return c.repair(r)
+	}
+	starting := action == "enable" || action == "recover"
+	if action == "recover" && (err != nil || !r.Enabled || r.Inventory != c.Expected || r.Lifecycle == nil) {
+		return errors.New("explicit enable approval required")
+	}
+	// A stopped approval referencing a completion must have the private,
+	// digest-bound proof; missing/corrupt artifacts are not completion.
+	if starting && r.Lifecycle != nil && r.Lifecycle.Completion != "" {
+		if _, e := completionRecord(c.Client, r); e != nil {
+			return e
+		}
+	}
+	var next *Lifecycle
+	if starting {
+		if c.Observe == nil {
+			return errTrust
+		}
+		binding, e := c.Observe(ctx, c.Expected)
+		if e != nil {
+			return e
+		}
+		if action == "recover" && (r.Lifecycle.Binding.Boot != binding.Boot || r.Lifecycle.Binding.Install != binding.Install) {
+			return errors.New("installation continuity changed; enable approval required")
+		}
+		next, err = newLifecycle(binding)
+		if err != nil {
+			return err
+		}
+		if err := createLifetime(c.Client, next); err != nil {
+			return err
+		}
+	}
+	// Preserve the owned identity before fencing lets the supervisor unlink it.
+	owned, ownedErr := readProcess(c.RunDir, c.Owner)
+	// Never hold policy.lock across Stop/ACK. Autonomous fencing shares only
+	// this short lock, so a controller holding control.lock cannot deadlock it.
+	if err := policyUpdate(c.Client, func(r *Record) error {
+		if action == "disable" {
+			r.Enabled = false
+		}
+		r.Residual = true
+		if r.Lifecycle != nil && r.Lifecycle.Phase != "stopped" {
+			r.Lifecycle.Phase = "fenced"
+			r.Lifecycle.Completion = ""
+			if r.Lifecycle.Reason == "" {
+				r.Lifecycle.Reason = "admin-stop"
+			}
+		}
+		return nil
+	}); err != nil {
+		return err
+	}
+	if ownedErr != nil && !errors.Is(ownedErr, fs.ErrNotExist) {
+		return ownedErr
+	}
+	if err := c.teardownFor(r, owned); err != nil {
+		return err
+	}
+	if err := c.commitTeardown(r, owned, next); err != nil {
+		return err
+	}
+	if starting {
+		if err := c.Processes.Start(c.Expected); err != nil {
+			return err
+		}
+		return finalizeStart(c.Client, next.Epoch)
+	}
+	return nil
+}
+
+// Keep the post-teardown policy commit separate from socket reconciliation.
+func (c Controller) commitTeardown(before Record, owned Process, next *Lifecycle) error {
+	observed, err := readRecord(c.StateDir, c.Owner)
+	if err != nil {
+		return err
+	}
+	if !sameTeardownEpoch(observed, before) {
+		return errTrust
+	}
+	published := ""
+	if observed.Lifecycle != nil {
+		published = observed.Lifecycle.Completion
+	}
+	// Rename makes completion visible before the owner's directory fsync and
+	// lock release. Wait only for that publisher; every other error is final.
+	ctx, cancel := context.WithTimeout(context.Background(), time.Second)
+	defer cancel()
+	last := errBusy
+	// A budget expiry keeps the last lock refusal; it never certifies success.
+	_ = waitHealthy(ctx, func(context.Context) bool {
+		last = policyUpdate(c.Client, func(r *Record) error {
+			if !sameTeardownEpoch(*r, before) || !runtimeAbsent(c.Client) {
+				return errTrust
+			}
+			if published != "" && (r.Lifecycle == nil || r.Lifecycle.Completion != published) {
+				return errTrust
+			}
+			if r.Lifecycle != nil && r.Lifecycle.Completion != "" {
+				proof, err := completionRecord(c.Client, *r)
+				if err != nil {
+					return err
+				}
+				if owned.PID > 1 {
+					if proof.Repair != nil || proof.Process != owned {
+						return errTrust
+					}
+				} else if before.Lifecycle == nil || before.Lifecycle.Phase != "stopped" {
+					return errTrust
+				}
+			}
+			if next != nil {
+				*r = Record{Inventory: c.Expected, Enabled: true, Residual: true, Lifecycle: next}
+			} else {
+				r.Residual = false
+				if r.Lifecycle != nil {
+					r.Lifecycle.Phase = "stopped"
+				}
+			}
+			return nil
+		})
+		return !errors.Is(last, errBusy)
+	})
+	return last
+}
+
+func (c Controller) teardown() error {
+	r, err := readRecord(c.StateDir, c.Owner)
+	if err != nil && !errors.Is(err, fs.ErrNotExist) {
+		return err
+	}
+	p, err := readProcess(c.RunDir, c.Owner)
+	if err != nil && !errors.Is(err, fs.ErrNotExist) {
+		return err
+	}
+	return c.teardownFor(r, p)
+}
+
+func (c Controller) teardownFor(before Record, owned Process) error {
+	r, err := readRecord(c.StateDir, c.Owner)
+	if err != nil && !errors.Is(err, fs.ErrNotExist) {
+		return err
+	}
+	if !sameTeardownEpoch(r, before) {
+		return errTrust
+	}
+	if r.Lifecycle != nil && r.Lifecycle.Completion != "" {
+		if owned.PID == 0 && before.Lifecycle != nil && before.Lifecycle.Phase != "stopped" {
+			return errTrust // no pre-fence process identity to reconcile this exit
+		}
+		if _, err := completionRecord(c.Client, r); err != nil {
+			return err
+		}
+	}
+	return c.teardownRecord(r, owned)
+}
+
+// Phase, completion and enabled/residual flags may change while the owning
+// supervisor exits; inventory, epoch, binding and optional lease must not.
+func sameTeardownEpoch(current, before Record) bool {
+	if current.Inventory != before.Inventory {
+		return false
+	}
+	a, b := current.Lifecycle, before.Lifecycle
+	if a == nil || b == nil {
+		return a == nil && b == nil
+	}
+	if a.Epoch != b.Epoch || a.Binding != b.Binding {
+		return false
+	}
+	if a.Lease == nil || b.Lease == nil {
+		return a.Lease == nil && b.Lease == nil
+	}
+	return *a.Lease == *b.Lease
+}
+
+// Reconcile runtime against the approval and pre-fence owned process witness.
+func (c Controller) teardownRecord(r Record, owned Process) error {
+	p, err := readProcess(c.RunDir, c.Owner)
+	if err != nil && !errors.Is(err, fs.ErrNotExist) {
+		return err
+	}
+	if err == nil {
+		if p != owned || p.Inventory.Package != c.Expected.Package {
+			return errTrust
+		}
+		if err := c.Processes.Stop(p); err != nil {
+			if c.waitTeardownCompletion(r, owned) == nil {
+				return nil
+			}
+			return err
+		}
+	} else if owned.PID > 1 {
+		// The owner can unlink after our approval snapshot but before this read.
+		// Absence alone is never success, including the artifact-before-commit gap.
+		if c.waitTeardownCompletion(r, owned) == nil {
+			return nil
+		}
+		return errTrust
+	}
+	if errors.Is(err, fs.ErrNotExist) && r.Lifecycle != nil && r.Lifecycle.Phase == "fenced" {
+		// Runtime unlink followed by a crash is not committed completion. Only
+		// a verified new boot + absent /run artifacts permits explicit repair.
+		if c.Boot == nil || !runtimeAbsent(c.Client) {
+			return errTrust
+		}
+		boot, e := c.Boot()
+		if e != nil || boot == "" || boot == r.Lifecycle.Binding.Boot {
+			return errTrust
+		}
+	}
+	if err := cleanRuntime(c.Client, err == nil); err != nil {
+		if c.waitTeardownCompletion(r, owned) == nil {
+			return nil
+		}
+		return err
+	}
+	return nil
+}
+
+func (c Controller) waitTeardownCompletion(before Record, owned Process) error {
+	if before.Lifecycle == nil || owned.PID <= 1 || owned.Inventory != before.Inventory || owned.Boot != before.Lifecycle.Binding.Boot {
+		return errTrust
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), time.Second)
+	defer cancel()
+	return waitHealthy(ctx, func(context.Context) bool {
+		r, err := readRecord(c.StateDir, c.Owner)
+		if err != nil || !sameTeardownEpoch(r, before) {
+			return false
+		}
+		proof, err := completionRecord(c.Client, r)
+		return err == nil && proof.Process == owned && runtimeAbsent(c.Client)
+	})
+}
+
+// cleanRuntime is shared by verified controller and owned supervisor cleanup.
+// The caller must first prove child teardown; it never signals a recorded PID.
+func cleanRuntime(c Client, hasProcess bool) error {
+	for _, name := range []string{"helper.sock", "supervisor.sock"} {
+		path := filepath.Join(c.RunDir, name)
+		st, e := os.Lstat(path)
+		if e == nil {
+			s, ok := st.Sys().(*syscall.Stat_t)
+			if !ok || s.Uid != c.Owner || st.Mode()&os.ModeSocket == 0 || st.Mode().Perm()&0002 != 0 {
+				return errTrust
+			}
+			conn, err := net.DialTimeout("unix", path, 100*time.Millisecond)
+			if err == nil {
+				conn.Close()
+				return errors.New("unidentified live socket")
+			}
+			if !errors.Is(err, syscall.ECONNREFUSED) && !errors.Is(err, fs.ErrNotExist) {
+				return err
+			}
+			if err := os.Remove(path); err != nil && !errors.Is(err, fs.ErrNotExist) {
+				return err
+			}
+		} else if !errors.Is(e, fs.ErrNotExist) {
+			return e
+		}
+	}
+	if hasProcess {
+		return os.Remove(filepath.Join(c.RunDir, "process.json"))
+	}
+	return nil
+}

package/templates/native-go/backend/internal/helper/control_test.go

@@ -0,0 +1,250 @@
+package helper
+
+import (
+	"context"
+	"errors"
+	"net"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+)
+
+func staleSocket(t *testing.T, p string) {
+	l, err := net.ListenUnix("unix", &net.UnixAddr{Name: p, Net: "unix"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	l.SetUnlinkOnClose(false)
+	l.Close()
+}
+
+type fakeProcesses struct {
+	current          Process
+	stopped, started int
+	stopErr          error
+	stopHook         func(Process) error
+}
+
+func (f *fakeProcesses) Stop(want Process) error {
+	if f.stopHook != nil {
+		return f.stopHook(want)
+	}
+	if f.stopErr != nil {
+		return f.stopErr
+	}
+	if want != f.current {
+		return errTrust
+	}
+	f.stopped++
+	return nil
+}
+func (f *fakeProcesses) Start(Inventory) error { f.started++; return nil }
+
+func controlFixture(t *testing.T) (Controller, *fakeProcesses) {
+	c := fixture(t)
+	p := Process{Inventory: c.Expected, Mode: "supervised", PID: 1234, Boot: "fixture-boot", Start: "100", Device: 1, Inode: 2}
+	f := &fakeProcesses{current: p}
+	ctl := Controller{Client: c, Processes: f, Observe: func(context.Context, Inventory) (Binding, error) { return testBinding(), nil }}
+	record(t, c, Record{Inventory: c.Expected, Enabled: true, Lifecycle: testLifecycle("running")})
+	if err := writeProcess(c.RunDir, p); err != nil {
+		t.Fatal(err)
+	}
+	return ctl, f
+}
+
+func TestTeardownAndRecoveryVerifyOwnedProcess(t *testing.T) {
+	for _, kind := range []string{"pid", "boot", "start", "inode", "digest", "package"} {
+		t.Run(kind, func(t *testing.T) {
+			c, f := controlFixture(t)
+			switch kind {
+			case "pid":
+				f.current.PID++
+			case "boot":
+				f.current.Boot = "new-boot"
+			case "start":
+				f.current.Start = "reused"
+			case "inode":
+				f.current.Inode++
+			case "digest":
+				f.current.Inventory.PayloadDigest = "foreign"
+			case "package":
+				f.current.Inventory.Package = "Other"
+			}
+			if c.Control(context.Background(), "recover") == nil || f.stopped != 0 || f.started != 0 {
+				t.Fatal("foreign process touched")
+			}
+			if _, err := os.Stat(filepath.Join(c.RunDir, "process.json")); err != nil {
+				t.Fatal("failure evidence removed")
+			}
+		})
+	}
+	c, f := controlFixture(t)
+	if err := c.Control(context.Background(), "teardown"); err != nil {
+		t.Fatal(err)
+	}
+	if f.stopped != 1 || f.started != 0 {
+		t.Fatal(f)
+	}
+	if _, err := os.Stat(filepath.Join(c.RunDir, "process.json")); !os.IsNotExist(err) {
+		t.Fatal(err)
+	}
+	if err := c.Control(context.Background(), "recover"); err != nil || f.started != 1 {
+		t.Fatalf("%v %+v", err, f)
+	}
+}
+
+func TestDisableIsDurableAndPrecedesFailedCleanup(t *testing.T) {
+	c, f := controlFixture(t)
+	f.stopErr = errors.New("fixture stop failed")
+	if c.Control(context.Background(), "disable") == nil {
+		t.Fatal("cleanup falsely succeeded")
+	}
+	r, err := readRecord(c.StateDir, c.Owner)
+	if err != nil || r.Enabled || !r.Residual {
+		t.Fatalf("%+v %v", r, err)
+	}
+	s := c.Status(context.Background())
+	if s.State != "not-enabled" || s.Cleanup != "required" {
+		t.Fatal(s)
+	}
+	if c.Control(context.Background(), "recover") == nil || f.started != 0 {
+		t.Fatal("old recovery resurrected disabled integration")
+	}
+	f.stopErr = nil
+	if err := c.Control(context.Background(), "disable"); err != nil {
+		t.Fatal(err)
+	}
+	r, _ = readRecord(c.StateDir, c.Owner)
+	if r.Enabled || r.Residual {
+		t.Fatal(r)
+	}
+	if err := c.Control(context.Background(), "enable"); err != nil || f.started != 1 {
+		t.Fatalf("%v %+v", err, f)
+	}
+}
+
+func TestMissingApprovalAndUnknownActionCannotStart(t *testing.T) {
+	c, f := controlFixture(t)
+	os.Remove(filepath.Join(c.StateDir, "approval.json"))
+	if c.Control(context.Background(), "recover") == nil {
+		t.Fatal("recover without approval")
+	}
+	if c.Control(context.Background(), "arbitrary-command") == nil || f.started != 0 {
+		t.Fatal("arbitrary action accepted")
+	}
+}
+
+func TestSocketResidueCannotUnlinkForeignFilesOrLiveListeners(t *testing.T) {
+	for _, kind := range []string{"file", "symlink", "live", "stale"} {
+		t.Run(kind, func(t *testing.T) {
+			c, f := controlFixture(t)
+			// This test covers migration cleanup of pre-M1g owned stale sockets.
+			// A schema-2 incomplete transaction is tested separately and cannot use
+			// absent runtime alone as completion proof.
+			record(t, c.Client, Record{Inventory: c.Expected, Enabled: true})
+			os.Remove(filepath.Join(c.RunDir, "process.json"))
+			p := filepath.Join(c.RunDir, "helper.sock")
+			switch kind {
+			case "file":
+				os.WriteFile(p, []byte("foreign"), 0600)
+			case "symlink":
+				os.Symlink("foreign", p)
+			case "live":
+				live(t, c.Client, c.Expected, nil)
+			case "stale":
+				staleSocket(t, p)
+			}
+			err := c.Control(context.Background(), "teardown")
+			if kind == "stale" {
+				if err != nil {
+					t.Fatal(err)
+				}
+				if _, err := os.Lstat(p); !os.IsNotExist(err) {
+					t.Fatal("stale socket not removed")
+				}
+			} else {
+				if err == nil {
+					t.Fatal("foreign residue accepted")
+				}
+				if _, err := os.Lstat(p); err != nil {
+					t.Fatal("foreign residue removed")
+				}
+			}
+			if f.stopped != 0 {
+				t.Fatal("unrecorded process signalled")
+			}
+		})
+	}
+}
+
+func TestLifecycleGateBlocksResidueAndAllowsCleanStoppedState(t *testing.T) {
+	c, _ := controlFixture(t)
+	if c.CheckStopped(context.Background()) == nil {
+		t.Fatal("process residue accepted")
+	}
+	if err := c.Control(context.Background(), "teardown"); err != nil {
+		t.Fatal(err)
+	}
+	if err := c.CheckStopped(context.Background()); err != nil {
+		t.Fatal(err)
+	}
+	record(t, c.Client, Record{Inventory: c.Expected, Enabled: false, Residual: true})
+	if c.CheckStopped(context.Background()) == nil {
+		t.Fatal("failed disable accepted")
+	}
+}
+
+func TestStartupProbesShareOneDeadline(t *testing.T) {
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Millisecond)
+	defer cancel()
+	err := waitHealthy(ctx, func(probeCtx context.Context) bool {
+		if probeCtx != ctx {
+			t.Fatal("probe got an independent deadline")
+		}
+		<-probeCtx.Done()
+		return false
+	})
+	if !errors.Is(err, context.DeadlineExceeded) {
+		t.Fatal(err)
+	}
+	if err := waitHealthy(context.Background(), func(context.Context) bool { return true }); err != nil {
+		t.Fatal(err)
+	}
+}
+
+func TestSupervisorSocketResidueAlsoBlocksLifecycle(t *testing.T) {
+	c, _ := controlFixture(t)
+	os.Remove(filepath.Join(c.RunDir, "process.json"))
+	p := filepath.Join(c.RunDir, "supervisor.sock")
+	l, err := net.ListenUnix("unix", &net.UnixAddr{Name: p, Net: "unix"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer l.Close()
+	if c.CheckStopped(context.Background()) == nil {
+		t.Fatal("unidentified supervisor accepted")
+	}
+	if c.Control(context.Background(), "teardown") == nil {
+		t.Fatal("live supervisor ignored")
+	}
+}
+
+func TestOnlySupervisedProcessRecordsAreReadable(t *testing.T) {
+	for _, mode := range []string{"", "supervised", "unknown"} {
+		c, f := controlFixture(t)
+		p := f.current
+		p.Mode = mode
+		if err := writeProcess(c.RunDir, p); err != nil {
+			t.Fatal(err)
+		}
+		got, err := readProcess(c.RunDir, c.Owner)
+		if mode != "supervised" {
+			if err == nil {
+				t.Fatal("unsupervised process mode accepted")
+			}
+		} else if err != nil || got != p {
+			t.Fatalf("mode %q: %+v %v", mode, got, err)
+		}
+	}
+}