controller: F7 atomic volume dumps + F6 no-single-copy + F5 stale-primary sweep (WIP, pre-build)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017CDMFpFx84pfviCTVuGGhf
2026-07-12 09:00:16 +02:00
parent b389e73fff
commit 68b3a3932e
13 changed files with 569 additions and 57 deletions
@@ -13,6 +13,7 @@ import (

 	"gitea.dooplex.hu/admin/felhom-controller/internal/config"
 	"gitea.dooplex.hu/admin/felhom-controller/internal/settings"
+	"gitea.dooplex.hu/admin/felhom-controller/internal/system"
 )

 // Manager orchestrates app-data backups: database dumps and Docker-volume tars.
@@ -44,6 +45,16 @@ type Manager struct {
 	// disconnected) can be unit-tested without Docker. Nil → the real DumpAppVolumesSafe.
 	dumpVolumesSafe func(stackName string) error

+	// F7 tar seam — the ONE docker exec inside DumpAppVolumes, overridable so the atomic-write
+	// behaviour (tmp+fsync+rename; the last good `.tar` survives a mid-write failure) is unit-testable
+	// without Docker. It must write the tar to `<dumpDir>/<volName>.tar.tmp` and return the combined
+	// output + error. Nil → the real `docker run … alpine tar cf …tar.tmp`.
+	tarVolume func(volName, dumpDir string) ([]byte, error)
+
+	// F6 per-app tier-2 seam — overridable so RunAllTier2's app SELECTION (now incl. volume-only apps)
+	// is unit-testable without rsync/du. Nil → the real RunTier2.
+	perAppTier2 func(stackName string) error
+
 	// generateSecret (O4), if set, produces a replacement value for a RESETTABLE secret that could
 	// not be recovered during restore-from-unit (wired to stacks.Manager.GenerateSecretForField in
 	// main.go). Nil / ok=false → the secret stays absent and the restore proceeds with a loud WARN.
@@ -95,8 +106,21 @@ type FullBackupStatus struct {
 	// Flash messages (set by handlers, passed through redirect)
 	FlashSuccess string
 	FlashError   string
+
+	// SingleCopyWarning (F6, CAMPAIGN-3) is a non-empty honest Hungarian notice when the box has NO
+	// off-drive target at all — tier-1 is then the ONLY local copy and 3-2-1 needs a 2nd drive or
+	// offsite. Empty when an off-drive (tier-2) target exists. Never a fake 3-2-1 guarantee.
+	SingleCopyWarning string
 }

+// systemDriveLabel is the human label for the internal SSD / system drive (F6 — a sys_drive app's
+// backup used to render with a blank drive label). Matches the tier-2 UI wording.
+const systemDriveLabel = "Belső SSD (rendszer)"
+
+// singleCopyNotice is the honest single-drive signal (F6): shown when no off-drive tier-2 target
+// exists, instead of silently implying a 3-2-1 guarantee the box cannot provide.
+const singleCopyNotice = "Csak egy másolat készül (nincs második meghajtó) — a 3-2-1 mentéshez csatlakoztasson egy második meghajtót vagy offsite tárolót."
+
 // DBDumpStatus holds the last DB dump result.
 type DBDumpStatus struct {
 	LastRun  time.Time
@@ -336,6 +360,11 @@ func (m *Manager) runDBDumpsInternal(ctx context.Context) error {
 	// Phase 2: refresh each deployed app's self-contained recovery unit (compose + manifest).
 	m.captureAllRecoveryUnits()

+	// F5 (CAMPAIGN-3): after the units are fresh on the CURRENT drives, prune any orphaned
+	// backups/primary/<app> dir an app left on an OLD drive when its HDD_PATH moved — pure disk
+	// residue, invisible in the snapshot list. Guarded (deployed + different current drive only).
+	m.pruneStalePrimaryDirs()
+
 	// No silent partials: a DB-dump or volume-dump failure fails the whole run.
 	if !allOK {
 		return fmt.Errorf("some backup steps failed: %s", strings.Join(failedSummaryLines(summary), "; "))
@@ -436,22 +465,35 @@ func (m *Manager) DumpAppVolumes(stackName string) error {
 	var dumpErrors []string
 	for _, volName := range volumes {
 		tarPath := filepath.Join(dumpDir, volName+".tar")
+		// F7 (CAMPAIGN-3, HIGH): write the tar to a `.tar.tmp` sibling and only atomically rename it
+		// over the restore point on success — the same crash-safe pattern the DB-dump path uses
+		// (appbackup/dbdump.go DumpOne). Before this, tar wrote the `.tar` IN PLACE, so a mid-write NFS
+		// cut left a 0-byte tar REPLACING the last good dump (tier-1 restore is replace-semantics → an
+		// empty volume). Now a failed/interrupted write only ever touches the `.tmp`; the last good
+		// `.tar` is untouched. The `.tmp` name (ends `.tmp`, not `.tar`) is invisible to the
+		// restore-point/stale scans, so it is never mistaken for a restore point.
+		tmpPath := tarPath + ".tmp"
 		if m.isDebug() {
 			m.logger.Printf("[DEBUG] [backup] Dumping volume %s for %s", volName, stackName)
 		}

-		ctx, cancel := context.WithTimeout(context.Background(), 10*time.Minute)
-		cmd := exec.CommandContext(ctx, "docker", "run", "--rm",
-			"-v", volName+":/vol:ro",
-			"-v", dumpDir+":/out",
-			"alpine", "tar", "cf", "/out/"+volName+".tar", "-C", "/vol", ".")
-		out, err := cmd.CombinedOutput()
-		cancel()
+		out, err := m.tarVolumeOrDefault(volName, dumpDir)

 		if err != nil {
-			m.logger.Printf("[WARN] [backup] Volume dump failed for %s/%s: %s — %v",
+			// Any tar error or context timeout (incl. a dead NFS target → EIO): remove ONLY the tmp,
+			// leave the existing `.tar` restore point byte-untouched, WARN, continue.
+			m.logger.Printf("[WARN] [backup] Volume dump failed for %s/%s (last good dump preserved): %s — %v",
 				stackName, volName, strings.TrimSpace(string(out)), err)
-			os.Remove(tarPath)
+			os.Remove(tmpPath)
+			dumpErrors = append(dumpErrors, volName)
+			continue
+		}
+
+		// fsync the tmp file (flush the tar to disk) then atomically rename over the restore point.
+		if err := atomicPromoteTar(tmpPath, tarPath); err != nil {
+			m.logger.Printf("[WARN] [backup] Volume dump promote failed for %s/%s (last good dump preserved): %v",
+				stackName, volName, err)
+			os.Remove(tmpPath)
 			dumpErrors = append(dumpErrors, volName)
 			continue
 		}
@@ -461,17 +503,24 @@ func (m *Manager) DumpAppVolumes(stackName string) error {
 		}
 	}

-	// Clean up tars for volumes that no longer exist
+	// Clean up tars (and any orphan `.tar.tmp` from a killed run) for volumes that no longer exist.
 	entries, _ := os.ReadDir(dumpDir)
 	activeVols := make(map[string]bool)
 	for _, v := range volumes {
 		activeVols[v+".tar"] = true
 	}
 	for _, e := range entries {
-		if !activeVols[e.Name()] && strings.HasSuffix(e.Name(), ".tar") {
-			os.Remove(filepath.Join(dumpDir, e.Name()))
+		name := e.Name()
+		// A leftover `.tar.tmp` is never a restore point — always safe to remove (its `.tar` sibling,
+		// if any, is the real restore point and is handled by the `.tar` branch).
+		if strings.HasSuffix(name, ".tar.tmp") {
+			os.Remove(filepath.Join(dumpDir, name))
+			continue
+		}
+		if !activeVols[name] && strings.HasSuffix(name, ".tar") {
+			os.Remove(filepath.Join(dumpDir, name))
 			if m.isDebug() {
-				m.logger.Printf("[DEBUG] [backup] Removed stale volume dump: %s/%s", stackName, e.Name())
+				m.logger.Printf("[DEBUG] [backup] Removed stale volume dump: %s/%s", stackName, name)
 			}
 		}
 	}
@@ -482,6 +531,51 @@ func (m *Manager) DumpAppVolumes(stackName string) error {
 	return nil
 }

+// tarVolumeOrDefault runs the F7 tar seam (m.tarVolume) or, when unset, the real docker tar into the
+// `<volName>.tar.tmp` sibling under dumpDir. The 10-minute bound matches the original.
+func (m *Manager) tarVolumeOrDefault(volName, dumpDir string) ([]byte, error) {
+	if m.tarVolume != nil {
+		return m.tarVolume(volName, dumpDir)
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Minute)
+	defer cancel()
+	cmd := exec.CommandContext(ctx, "docker", "run", "--rm",
+		"-v", volName+":/vol:ro",
+		"-v", dumpDir+":/out",
+		"alpine", "tar", "cf", "/out/"+volName+".tar.tmp", "-C", "/vol", ".")
+	return cmd.CombinedOutput()
+}
+
+// atomicPromoteTar fsyncs a completed `.tar.tmp` then atomically renames it over the final `.tar`
+// (same-dir rename = atomic on the target fs). It mirrors DumpOne's crash-safety (F7) and EXCEEDS it
+// by also best-effort fsync'ing the directory entry, so the rename itself survives a power loss — the
+// DB-dump path fsyncs the file but not the dir; a follow-up could add the dir fsync there too. On any
+// error the tmp is left for the caller to remove; the final `.tar` is never touched here except by a
+// successful rename.
+func atomicPromoteTar(tmpPath, finalPath string) error {
+	f, err := os.Open(tmpPath)
+	if err != nil {
+		return fmt.Errorf("opening tmp dump: %w", err)
+	}
+	if err := f.Sync(); err != nil {
+		f.Close()
+		return fmt.Errorf("syncing tmp dump: %w", err)
+	}
+	if err := f.Close(); err != nil {
+		return fmt.Errorf("closing tmp dump: %w", err)
+	}
+	if err := os.Rename(tmpPath, finalPath); err != nil {
+		return fmt.Errorf("renaming tmp dump: %w", err)
+	}
+	// Best-effort: fsync the directory so the rename is durable (ignore errors — the rename already
+	// made the new content visible; this only hardens against a power loss immediately after).
+	if dir, derr := os.Open(filepath.Dir(finalPath)); derr == nil {
+		_ = dir.Sync()
+		_ = dir.Close()
+	}
+	return nil
+}
+
 // DumpAppVolumesSafe stops the stack before dumping volumes and restarts after.
 // Prevents inconsistent tars of live database volumes (e.g. PostgreSQL).
 // Protected stacks that reject StopStack will return an error — callers handle as warning.
@@ -749,6 +843,7 @@ func (m *Manager) GetFullStatus(nextDBDump time.Time) *FullBackupStatus {
 		// Update dynamic fields that don't need subprocess calls
 		status.Running = m.running
 		status.NextDBDump = nextDBDump
+		status.SingleCopyWarning = m.singleCopyWarning() // F6: honest single-drive signal
 		// Deep-copy lastDBDump so callers cannot mutate shared state.
 		if m.lastDBDump != nil {
 			copyDump := *m.lastDBDump
@@ -786,10 +881,11 @@ func (m *Manager) GetFullStatus(nextDBDump time.Time) *FullBackupStatus {

 	// No cache yet — return a minimal status (first page load before cache is populated)
 	status := &FullBackupStatus{
-		Enabled:        m.cfg.Backup.Enabled,
-		Running:        m.running,
-		DBDumpSchedule: m.cfg.Backup.DBDumpSchedule,
-		NextDBDump:     nextDBDump,
+		Enabled:           m.cfg.Backup.Enabled,
+		Running:           m.running,
+		DBDumpSchedule:    m.cfg.Backup.DBDumpSchedule,
+		NextDBDump:        nextDBDump,
+		SingleCopyWarning: m.singleCopyWarning(), // F6
 	}
 	if m.lastDBDump != nil {
 		copyDump := *m.lastDBDump
@@ -802,6 +898,114 @@ func (m *Manager) GetFullStatus(nextDBDump time.Time) *FullBackupStatus {
 	return status
 }

+// hasOffDriveTarget reports whether any registered, schedulable storage path lives on a physical disk
+// OTHER than the system drive — i.e. whether a genuine off-drive (tier-2) copy is possible at all.
+// When false the box is single-drive: tier-1 is the ONLY local copy and 3-2-1 needs a 2nd drive or
+// offsite (F6 — surfaced honestly via SingleCopyWarning, never a faked guarantee).
+func (m *Manager) hasOffDriveTarget() bool {
+	if m.settings == nil || m.systemDataPath == "" {
+		return false
+	}
+	for _, sp := range m.settings.GetSchedulableStoragePaths() {
+		if sp.Path == m.systemDataPath || system.SamePhysicalDevice(m.systemDataPath, sp.Path) {
+			continue
+		}
+		return true
+	}
+	return false
+}
+
+// singleCopyWarning returns the honest single-drive notice, or "" when an off-drive target exists.
+func (m *Manager) singleCopyWarning() string {
+	if m.hasOffDriveTarget() {
+		return ""
+	}
+	return singleCopyNotice
+}
+
+// sysDriveLabelFor returns the drive label for a stack's tier-1 restore point — the clear
+// system-drive label for a sys_drive (volume-only) app (F6: never blank), else the enrolled drive's
+// storage label.
+func (m *Manager) sysDriveLabelFor(stackName string) string {
+	drive := m.GetAppDrivePath(stackName)
+	if drive == "" {
+		return ""
+	}
+	if drive == m.systemDataPath {
+		return systemDriveLabel
+	}
+	if m.settings != nil {
+		return m.settings.GetStorageLabel(drive)
+	}
+	return ""
+}
+
+// pruneStalePrimaryDirs removes orphaned `backups/primary/<app>` dirs left on a drive after an app's
+// HDD_PATH moved to another drive (F5, CAMPAIGN-3 — pure disk residue, invisible in the snapshot
+// list). LOAD-BEARING GUARDS: a dir is removed ONLY when <app> is currently deployed AND its current
+// namespace root differs from this dir's drive. It NEVER removes the dir on the app's CURRENT drive
+// (that IS the live restore point), and NEVER removes a dir for an app NOT in the deployed set (an
+// undeployed app's last backup is still its restore point — orphaned-app cleanup is a separate,
+// user-driven concern). Only operates strictly under a `backups/primary/` prefix.
+func (m *Manager) pruneStalePrimaryDirs() {
+	if m.stackProvider == nil {
+		return
+	}
+	// Current namespace root per DEPLOYED app.
+	current := map[string]string{}
+	for _, s := range m.stackProvider.ListDeployedStacks() {
+		if drive := m.GetAppDrivePath(s.Name); drive != "" {
+			current[s.Name] = filepath.Clean(m.namespaceRoot(drive))
+		}
+	}
+	// Candidate drives to scan: the system drive + every registered storage path.
+	var nsRoots []string
+	if m.systemDataPath != "" {
+		nsRoots = append(nsRoots, filepath.Clean(m.namespaceRoot(m.systemDataPath)))
+	}
+	if m.settings != nil {
+		for _, sp := range m.settings.GetStoragePaths() {
+			nsRoots = append(nsRoots, filepath.Clean(NamespaceRoot(sp.Path, true)))
+		}
+	}
+	seen := map[string]bool{}
+	for _, nsRoot := range nsRoots {
+		if seen[nsRoot] {
+			continue
+		}
+		seen[nsRoot] = true
+		primaryDir := PrimaryBackupPath(nsRoot)
+		entries, err := os.ReadDir(primaryDir)
+		if err != nil {
+			continue // absent/unreadable (e.g. a disconnected drive) — nothing to prune here
+		}
+		for _, e := range entries {
+			if !e.IsDir() {
+				continue
+			}
+			app := e.Name()
+			cur, deployed := current[app]
+			if !deployed {
+				continue // GUARD: an undeployed app's last backup is still its restore point
+			}
+			if cur == nsRoot {
+				continue // GUARD: this IS the app's current drive — the live restore point
+			}
+			stalePath := RecoveryUnitPath(nsRoot, app)
+			// Prefix safety: only ever remove strictly inside `backups/primary/` (no surprise user data).
+			if !strings.HasPrefix(filepath.Clean(stalePath)+string(filepath.Separator),
+				filepath.Clean(primaryDir)+string(filepath.Separator)) {
+				continue
+			}
+			if err := os.RemoveAll(stalePath); err != nil {
+				m.logger.Printf("[WARN] [backup] F5: could not remove stale primary dir for %s on old drive: %v", app, err)
+			} else {
+				m.logger.Printf("[INFO] [backup] F5: removed stale primary backup dir for %s on an old drive (app now on %s)", app, cur)
+			}
+		}
+	}
+}
+
 // isDebug returns true if logging level is "debug".
 func (m *Manager) isDebug() bool {
 	return m.cfg != nil && m.cfg.Logging.Level == "debug"
@@ -0,0 +1,107 @@
+package backup
+
+import (
+	"io"
+	"log"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/config"
+	"gitea.dooplex.hu/admin/felhom-controller/internal/settings"
+)
+
+// f5Manager wires a Manager with a real settings store holding the given enrolled drives, plus a
+// sys drive, and a fake provider mapping each deployed app to its CURRENT HDD path.
+func f5Manager(t *testing.T, sysDrive string, enrolled []string, deployed map[string]string) *Manager {
+	t.Helper()
+	sett, err := settings.Load(filepath.Join(t.TempDir(), "settings.json"), log.New(io.Discard, "", 0))
+	if err != nil {
+		t.Fatal(err)
+	}
+	for _, d := range enrolled {
+		if err := sett.AddStoragePath(settings.StoragePath{Path: d, Label: filepath.Base(d), Schedulable: true}); err != nil {
+			t.Fatal(err)
+		}
+	}
+	var stacks []StackSummary
+	for name := range deployed {
+		stacks = append(stacks, StackSummary{Name: name})
+	}
+	cfg := &config.Config{}
+	cfg.Paths.SystemDataPath = sysDrive
+	fake := &volDumpFakeProvider{stacks: stacks, hdd: deployed}
+	return &Manager{cfg: cfg, settings: sett, logger: log.New(io.Discard, "", 0),
+		systemDataPath: sysDrive, stackProvider: fake}
+}
+
+// seedPrimaryDir creates a backups/primary/<app> dir (with a marker file) under a drive's namespace.
+func seedPrimaryDir(t *testing.T, driveNsRoot, app string) string {
+	t.Helper()
+	dir := RecoveryUnitPath(driveNsRoot, app)
+	if err := os.MkdirAll(filepath.Join(dir, "volume-dumps"), 0755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(dir, "manifest.json"), []byte("{}"), 0644); err != nil {
+		t.Fatal(err)
+	}
+	return dir
+}
+
+// F5: an app redeployed from drive A to drive B → its stale primary dir on A is removed, its live
+// dir on B is kept. Guard red-proof lives in the two "kept" assertions.
+func TestPruneStalePrimaryDirs_RemovesRedeployedResidue(t *testing.T) {
+	driveA := t.TempDir()
+	driveB := t.TempDir()
+	sys := filepath.Join(t.TempDir(), "sys")
+	// nextcloud currently on B; a stale unit remains on A.
+	m := f5Manager(t, sys, []string{driveA, driveB}, map[string]string{"nextcloud": driveB})
+
+	nsA := NamespaceRoot(driveA, true)
+	nsB := NamespaceRoot(driveB, true)
+	staleA := seedPrimaryDir(t, nsA, "nextcloud")
+	liveB := seedPrimaryDir(t, nsB, "nextcloud")
+
+	m.pruneStalePrimaryDirs()
+
+	if _, err := os.Stat(staleA); !os.IsNotExist(err) {
+		t.Errorf("stale primary dir on the OLD drive was not removed (F5)")
+	}
+	if _, err := os.Stat(liveB); err != nil {
+		t.Errorf("the LIVE primary dir on the current drive was wrongly removed: %v", err)
+	}
+}
+
+// GUARD: an UNDEPLOYED app's primary dir is a restore point — it must be KEPT (companion: drop the
+// `!deployed { continue }` guard → this dir is deleted → fail).
+func TestPruneStalePrimaryDirs_KeepsUndeployedRestorePoint(t *testing.T) {
+	driveA := t.TempDir()
+	sys := filepath.Join(t.TempDir(), "sys")
+	// Only "nextcloud" is deployed (on A); "removed-app" has a leftover unit but is NOT deployed.
+	m := f5Manager(t, sys, []string{driveA}, map[string]string{"nextcloud": driveA})
+	nsA := NamespaceRoot(driveA, true)
+	seedPrimaryDir(t, nsA, "nextcloud")
+	orphan := seedPrimaryDir(t, nsA, "removed-app")
+
+	m.pruneStalePrimaryDirs()
+
+	if _, err := os.Stat(orphan); err != nil {
+		t.Errorf("an UNDEPLOYED app's restore point was wrongly deleted (guard failure): %v", err)
+	}
+}
+
+// GUARD: the dir on an app's CURRENT drive is never touched (already covered above, but assert it
+// directly with a single-drive app so no cross-drive move is involved).
+func TestPruneStalePrimaryDirs_KeepsCurrentDriveDir(t *testing.T) {
+	driveA := t.TempDir()
+	sys := filepath.Join(t.TempDir(), "sys")
+	m := f5Manager(t, sys, []string{driveA}, map[string]string{"nextcloud": driveA})
+	nsA := NamespaceRoot(driveA, true)
+	live := seedPrimaryDir(t, nsA, "nextcloud")
+
+	m.pruneStalePrimaryDirs()
+
+	if _, err := os.Stat(live); err != nil {
+		t.Errorf("the current-drive primary dir was wrongly removed: %v", err)
+	}
+}
@@ -0,0 +1,76 @@
+package backup
+
+import (
+	"io"
+	"log"
+	"path/filepath"
+	"testing"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/config"
+	"gitea.dooplex.hu/admin/felhom-controller/internal/settings"
+)
+
+// F6: a volume-only app (no HDD_PATH → data on sys_drive) must now flow through the tier-2 run — it
+// used to be skipped, leaving a single controller-level copy. COMPANION red-proof: restore the
+// `GetStackHDDPath == "" { continue }` skip in RunAllTier2 → "actualbudget" is absent → this fails.
+func TestRunAllTier2_IncludesVolumeOnlyApps(t *testing.T) {
+	cfg := &config.Config{}
+	cfg.Paths.SystemDataPath = filepath.Join(t.TempDir(), "sys")
+	fake := &volDumpFakeProvider{
+		stacks: []StackSummary{{Name: "nextcloud"}, {Name: "actualbudget"}},
+		hdd:    map[string]string{"nextcloud": filepath.Join(t.TempDir(), "usb")}, // actualbudget = volume-only
+	}
+	m := &Manager{cfg: cfg, logger: log.New(io.Discard, "", 0),
+		systemDataPath: cfg.Paths.SystemDataPath, stackProvider: fake}
+
+	var processed []string
+	m.perAppTier2 = func(name string) error { processed = append(processed, name); return nil }
+	m.RunAllTier2()
+
+	var sawVolumeOnly bool
+	for _, p := range processed {
+		if p == "actualbudget" {
+			sawVolumeOnly = true
+		}
+	}
+	if !sawVolumeOnly {
+		t.Errorf("volume-only app 'actualbudget' was NOT included in the tier-2 run (F6): processed=%v", processed)
+	}
+}
+
+// F6: a single-drive box (no off-drive target) surfaces the honest single-copy notice; a box with an
+// off-drive target does not (never a faked 3-2-1 guarantee).
+func TestSingleCopyWarning_HonestOnSingleDrive(t *testing.T) {
+	// Single-drive: only the system drive, no schedulable off-drive paths.
+	m1, _ := newTestManager(t, "/srv/sys")
+	if got := m1.singleCopyWarning(); got != singleCopyNotice {
+		t.Errorf("single-drive box must surface the honest notice, got %q", got)
+	}
+
+	// Multi-drive: an enrolled off-drive path → no warning.
+	m2, sett := newTestManager(t, "/srv/sys")
+	if err := sett.AddStoragePath(settings.StoragePath{Path: "/mnt/usb", Label: "USB", Schedulable: true}); err != nil {
+		t.Fatal(err)
+	}
+	if got := m2.singleCopyWarning(); got != "" {
+		t.Errorf("a box with an off-drive target must NOT show the single-copy notice, got %q", got)
+	}
+}
+
+// F6: a sys_drive (volume-only) app's restore point carries a clear drive label, never blank.
+func TestRestorePoint_SysDriveLabelNotBlank(t *testing.T) {
+	cfg := &config.Config{}
+	sysDrive := t.TempDir()
+	cfg.Paths.SystemDataPath = sysDrive
+	fake := &volDumpFakeProvider{
+		stacks:  []StackSummary{{Name: "actualbudget"}},
+		volumes: map[string][]string{"actualbudget": {"actualbudget_data"}},
+	}
+	m := &Manager{cfg: cfg, logger: log.New(io.Discard, "", 0),
+		systemDataPath: sysDrive, stackProvider: fake}
+
+	label := m.sysDriveLabelFor("actualbudget")
+	if label != systemDriveLabel {
+		t.Errorf("sys_drive app drive label = %q, want %q (never blank — F6)", label, systemDriveLabel)
+	}
+}
@@ -47,8 +47,10 @@ func defaultOffboxRunner(ctx context.Context, env []string, args ...string) ([]b
 }

 // SetOffboxRunner overrides the restic exec (tests). SetOffboxNotify wires the failure→operator alert.
-func (m *Manager) SetOffboxRunner(r offboxRunner)                              { m.offboxRunner = r }
-func (m *Manager) SetOffboxNotify(fn func(dur time.Duration, snapshots int, err error)) { m.offboxNotify = fn }
+func (m *Manager) SetOffboxRunner(r offboxRunner) { m.offboxRunner = r }
+func (m *Manager) SetOffboxNotify(fn func(dur time.Duration, snapshots int, err error)) {
+	m.offboxNotify = fn
+}

 func (m *Manager) runner() offboxRunner {
 	if m.offboxRunner != nil {
@@ -57,10 +59,10 @@ func (m *Manager) runner() offboxRunner {
 	return defaultOffboxRunner
 }

-func (m *Manager) offboxDir() string         { return filepath.Join(m.cfg.Paths.DataDir, "offbox") }
-func (m *Manager) offboxKeyPath() string     { return filepath.Join(m.offboxDir(), "ssh_key") }
-func (m *Manager) offboxPwPath() string      { return filepath.Join(m.offboxDir(), "repo_password") }
-func (m *Manager) offboxKnownHosts() string  { return filepath.Join(m.offboxDir(), "known_hosts") }
+func (m *Manager) offboxDir() string        { return filepath.Join(m.cfg.Paths.DataDir, "offbox") }
+func (m *Manager) offboxKeyPath() string    { return filepath.Join(m.offboxDir(), "ssh_key") }
+func (m *Manager) offboxPwPath() string     { return filepath.Join(m.offboxDir(), "repo_password") }
+func (m *Manager) offboxKnownHosts() string { return filepath.Join(m.offboxDir(), "known_hosts") }

 // WriteOffboxSecrets persists the SSH private key + (auto-generated if empty) repo password + the pinned
 // known-host line as 0600/0644 files in the data dir. The key is provided out-of-band by the operator
@@ -93,7 +93,7 @@ func TestOffbox_BaseArgsCarryConnectTimeout(t *testing.T) {
 	base, env := m.offboxBaseArgs(sett.GetOffboxTarget())
 	joined := strings.Join(base, " ")
 	for _, want := range []string{
-		"-oConnectTimeout=10",        // THE spike Q8 fail-fast knob
+		"-oConnectTimeout=10",         // THE spike Q8 fail-fast knob
 		"-oStrictHostKeyChecking=yes", // no blind TOFU
 		"-oUserKnownHostsFile=",       // pinned host key
 		"-oBatchMode=yes",             // no interactive hang
@@ -466,8 +466,8 @@ const resticLockErr = "unable to create lock in backend: repository is already l
 // lockRunner records calls and returns the lock error for `backup` the first `lockTimes` times it's called.
 type lockRunner struct {
 	backups, prunes, unlockStale, unlockRemoveAll int
-	lockTimes                                      int
-	seq                                            []string
+	lockTimes                                     int
+	seq                                           []string
 }

 func (r *lockRunner) run(_ context.Context, _ []string, args ...string) ([]byte, error) {
@@ -629,10 +629,10 @@ func TestOffbox_ValidateRejectsInjection(t *testing.T) {
 		{Host: "-oProxyCommand=touch /tmp/pwn", User: "felhom", RepoPath: "/srv/repo"}, // ssh option injection
 		{Host: "nas;rm -rf /", User: "felhom", RepoPath: "/srv/repo"},                  // metacharacters
 		{Host: "nas.local", User: "-oProxyCommand=x", RepoPath: "/srv/repo"},           // user option injection
-		{Host: "nas.local", User: "felhom", RepoPath: "/srv/repo; evil"},              // path metacharacters
-		{Host: "nas.local", User: "felhom", RepoPath: "/srv/../etc"},                  // traversal
-		{Host: "nas local", User: "felhom", RepoPath: "/srv/repo"},                    // space
-		{Host: "nas.local", User: "felhom", RepoPath: "relative/path"},                // non-absolute
+		{Host: "nas.local", User: "felhom", RepoPath: "/srv/repo; evil"},               // path metacharacters
+		{Host: "nas.local", User: "felhom", RepoPath: "/srv/../etc"},                   // traversal
+		{Host: "nas local", User: "felhom", RepoPath: "/srv/repo"},                     // space
+		{Host: "nas.local", User: "felhom", RepoPath: "relative/path"},                 // non-absolute
 	}
 	for i, b := range bad {
 		bb := b
@@ -11,7 +11,7 @@ import "time"

 // RestoreOpResult is the terminal record of the most recent restore op.
 type RestoreOpResult struct {
-	Op         string    `json:"op"`      // "restore" | "tier2-restore" | "offbox-restore"
+	Op         string    `json:"op"` // "restore" | "tier2-restore" | "offbox-restore"
 	Stack      string    `json:"stack"`
 	OK         bool      `json:"ok"`
 	Message    string    `json:"message"`
@@ -34,9 +34,9 @@ type RecoveryManifest struct {
 	DisplayName    string            `json:"display_name"`
 	ControllerVer  string            `json:"controller_version"`
 	CreatedAt      string            `json:"created_at"`
-	Drive          string            `json:"drive"`          // HDD_PATH (in-guest mount)
-	NamespaceRoot  string            `json:"namespace_root"` // resolved felhom-data namespace root
-	ImagePins      []string          `json:"image_pins"`     // image NOT stored — re-pulled on restore
+	Drive          string            `json:"drive"`             // HDD_PATH (in-guest mount)
+	NamespaceRoot  string            `json:"namespace_root"`    // resolved felhom-data namespace root
+	ImagePins      []string          `json:"image_pins"`        // image NOT stored — re-pulled on restore
 	SecretEnvVars  []string          `json:"secret_env_vars"`   // NAMES only — recovered from guest/PBS
 	DataKeyEnvVars []string          `json:"data_key_env_vars"` // fail-closed gate on restore
 	SecretSource   string            `json:"secret_source"`     // human note: where secrets come from
@@ -24,13 +24,13 @@ type fakeRecoveryProvider struct {
 func (f *fakeRecoveryProvider) GetStackComposePath(string) (string, bool) {
 	return filepath.Join(f.info.StackDir, "docker-compose.yml"), true
 }
-func (f *fakeRecoveryProvider) ListDeployedStacks() []StackSummary       { return nil }
-func (f *fakeRecoveryProvider) GetStackHDDMounts(string) []string        { return nil }
-func (f *fakeRecoveryProvider) GetStackHDDPath(string) string            { return f.hdd }
-func (f *fakeRecoveryProvider) GetDockerVolumes(string) []string         { return nil }
-func (f *fakeRecoveryProvider) StopStack(string) error                   { f.stopped = true; return nil }
-func (f *fakeRecoveryProvider) StartStack(string) error                  { return nil }
-func (f *fakeRecoveryProvider) RefreshAndIsRunning(string) bool          { return f.running }
+func (f *fakeRecoveryProvider) ListDeployedStacks() []StackSummary { return nil }
+func (f *fakeRecoveryProvider) GetStackHDDMounts(string) []string  { return nil }
+func (f *fakeRecoveryProvider) GetStackHDDPath(string) string      { return f.hdd }
+func (f *fakeRecoveryProvider) GetDockerVolumes(string) []string   { return nil }
+func (f *fakeRecoveryProvider) StopStack(string) error             { f.stopped = true; return nil }
+func (f *fakeRecoveryProvider) StartStack(string) error            { return nil }
+func (f *fakeRecoveryProvider) RefreshAndIsRunning(string) bool    { return f.running }
 func (f *fakeRecoveryProvider) GetStackRecoveryInfo(string) (RecoveryInfo, bool) {
 	return f.info, true
 }
@@ -58,16 +58,11 @@ func (m *Manager) ListRestorePoints(stackName string) (points []RestorePoint, fo
 	newest = newestArtifact(AppDBDumpPath(nsRoot, stackName), ".sql", newest)
 	newest = newestArtifact(AppVolumeDumpPath(nsRoot, stackName), ".tar", newest)

-	driveLabel := ""
-	if drive := m.GetAppDrivePath(stackName); drive != "" && drive != m.systemDataPath && m.settings != nil {
-		driveLabel = m.settings.GetStorageLabel(drive)
-	}
-
 	return []RestorePoint{{
 		Time:       newest.UTC().Format(time.RFC3339),
 		ShortID:    restorePointShortID,
 		Tier:       1,
-		DriveLabel: driveLabel,
+		DriveLabel: m.sysDriveLabelFor(stackName),
 	}}, true
 }

@@ -206,19 +206,25 @@ func (m *Manager) RunAllTier2() {
 	}
 	var n int
 	for _, stack := range m.stackProvider.ListDeployedStacks() {
-		if m.stackProvider.GetStackHDDPath(stack.Name) == "" {
-			continue // not an HDD app — its data is on the rootfs, covered by PBS
-		}
+		// F6 (CAMPAIGN-3): volume-only apps (no HDD_PATH, backups on sys_drive) previously got NO
+		// tier-2 copy — a single controller-level copy on one device. They now flow through too: their
+		// recovery unit (which holds the db/volume dumps) gets a cross-drive second copy like any HDD
+		// app. Apps with no recovery unit yet (infra/never-backed-up) are a cheap no-op inside RunTier2
+		// (the `os.Stat(unitDir)` guard), so this doesn't spuriously copy protected stacks.
 		if m.settings != nil && (m.settings.IsDisconnected(m.GetAppDrivePath(stack.Name)) ||
 			m.settings.IsDecommissioned(m.GetAppDrivePath(stack.Name))) {
 			continue
 		}
-		if err := m.RunTier2(stack.Name); err != nil {
+		runOne := m.perAppTier2
+		if runOne == nil {
+			runOne = m.RunTier2
+		}
+		if err := runOne(stack.Name); err != nil {
 			m.logger.Printf("[WARN] [backup] Tier 2 failed for %s: %v", stack.Name, err)
 		}
 		n++
 	}
-	m.logger.Printf("[INFO] [backup] Tier 2 run complete: %d HDD app(s) processed", n)
+	m.logger.Printf("[INFO] [backup] Tier 2 run complete: %d app(s) processed (incl. volume-only — F6)", n)
 }

 // --- per-app config-panel view (drives the Tier-2 "Beállítás" page) ---
@@ -0,0 +1,119 @@
+package backup
+
+import (
+	"io"
+	"log"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/config"
+)
+
+// atomicTestManager wires a Manager whose GetAppDrivePath resolves to a single drive dir, with the
+// F7 tar seam injected. Returns the manager + the volume-dump dir where `.tar`/`.tar.tmp` land.
+func atomicTestManager(t *testing.T, stack, volName string, tar func(volName, dumpDir string) ([]byte, error)) (*Manager, string) {
+	t.Helper()
+	drive := t.TempDir()
+	cfg := &config.Config{}
+	cfg.Paths.SystemDataPath = drive
+	fake := &volDumpFakeProvider{
+		stacks:  []StackSummary{{Name: stack}},
+		volumes: map[string][]string{stack: {volName}},
+	}
+	m := &Manager{cfg: cfg, logger: log.New(io.Discard, "", 0),
+		systemDataPath: drive, stackProvider: fake, tarVolume: tar}
+	dumpDir := AppVolumeDumpPath(m.namespaceRoot(drive), stack)
+	return m, dumpDir
+}
+
+// Happy path: the tar seam writes a good `.tar.tmp`; DumpAppVolumes promotes it to `.tar`, leaves no
+// `.tmp`, and the content is exactly what the seam wrote.
+func TestDumpAppVolumes_HappyAtomicPromote(t *testing.T) {
+	const vol = "app_data"
+	want := []byte("GOOD-TAR-CONTENT")
+	m, dumpDir := atomicTestManager(t, "app", vol, func(_, dir string) ([]byte, error) {
+		return nil, os.WriteFile(filepath.Join(dir, vol+".tar.tmp"), want, 0644)
+	})
+
+	if err := m.DumpAppVolumes("app"); err != nil {
+		t.Fatalf("DumpAppVolumes: %v", err)
+	}
+	got, err := os.ReadFile(filepath.Join(dumpDir, vol+".tar"))
+	if err != nil {
+		t.Fatalf("final .tar missing: %v", err)
+	}
+	if string(got) != string(want) {
+		t.Errorf("final .tar content = %q, want %q", got, want)
+	}
+	if _, err := os.Stat(filepath.Join(dumpDir, vol+".tar.tmp")); !os.IsNotExist(err) {
+		t.Errorf("a `.tar.tmp` was left behind after a successful dump")
+	}
+}
+
+// THE F7 RED-PROOF: a good `.tar` already exists (the last restore point). The tar seam fails
+// mid-write (writes a partial/0-byte `.tar.tmp` then errors, simulating a NFS cut). The ORIGINAL
+// `.tar` MUST be byte-identical afterwards, and no 0-byte `.tar` may exist. Companion: revert
+// DumpAppVolumes to the in-place `tar cf …/<vol>.tar` write → the marker is truncated → this fails.
+func TestDumpAppVolumes_MidWriteFailurePreservesLastGood(t *testing.T) {
+	const vol = "app_data"
+	marker := []byte("LAST-GOOD-247-BYTE-DUMP-MARKER") // the campaign's 247-byte last-good dump, in spirit
+
+	m, dumpDir := atomicTestManager(t, "app", vol, func(_, dir string) ([]byte, error) {
+		// Simulate a mid-write cut: partially write the tmp, then fail (as a dead NFS target would).
+		_ = os.WriteFile(filepath.Join(dir, vol+".tar.tmp"), []byte("PARTIAL-TRUNCATED"), 0644)
+		return []byte("tar: write error: Input/output error"), errTest
+	})
+	// Pre-seed the last good restore point.
+	if err := os.MkdirAll(dumpDir, 0755); err != nil {
+		t.Fatal(err)
+	}
+	tarPath := filepath.Join(dumpDir, vol+".tar")
+	if err := os.WriteFile(tarPath, marker, 0644); err != nil {
+		t.Fatal(err)
+	}
+
+	err := m.DumpAppVolumes("app")
+	if err == nil {
+		t.Fatal("DumpAppVolumes must report failure when the tar step fails")
+	}
+
+	// The restore point is byte-identical — F7's whole point.
+	got, rerr := os.ReadFile(tarPath)
+	if rerr != nil {
+		t.Fatalf("the last good `.tar` was DESTROYED by a failed write (F7 regression): %v", rerr)
+	}
+	if string(got) != string(marker) {
+		t.Errorf("last good `.tar` was mutated by a failed write: got %q, want %q (F7 regression)", got, marker)
+	}
+	// No stray tmp and no 0-byte tar.
+	if _, err := os.Stat(filepath.Join(dumpDir, vol+".tar.tmp")); !os.IsNotExist(err) {
+		t.Errorf("the failed `.tar.tmp` was not cleaned up")
+	}
+	if info, _ := os.Stat(tarPath); info != nil && info.Size() == 0 {
+		t.Errorf("the restore point is now a 0-byte file (the exact F7 failure)")
+	}
+}
+
+// A leftover `.tar.tmp` from a previously-killed run is cleaned up by the next dump and is never a
+// restore point (its name ends `.tmp`, invisible to the `.tar` restore-point scan).
+func TestDumpAppVolumes_LeftoverTmpCleaned(t *testing.T) {
+	const vol = "app_data"
+	m, dumpDir := atomicTestManager(t, "app", vol, func(_, dir string) ([]byte, error) {
+		return nil, os.WriteFile(filepath.Join(dir, vol+".tar.tmp"), []byte("new"), 0644)
+	})
+	if err := os.MkdirAll(dumpDir, 0755); err != nil {
+		t.Fatal(err)
+	}
+	// A stale tmp for a DIFFERENT (removed) volume — must be swept.
+	stale := filepath.Join(dumpDir, "old_removed_vol.tar.tmp")
+	if err := os.WriteFile(stale, []byte("stale"), 0644); err != nil {
+		t.Fatal(err)
+	}
+	if err := m.DumpAppVolumes("app"); err != nil {
+		t.Fatalf("DumpAppVolumes: %v", err)
+	}
+	if _, err := os.Stat(stale); !os.IsNotExist(err) {
+		t.Errorf("a stale `.tar.tmp` was left behind (must be swept)")
+	}
+}
@@ -30,8 +30,8 @@ func (f *volDumpFakeProvider) StopStack(name string) error {
 	f.stopped = append(f.stopped, name)
 	return nil
 }
-func (f *volDumpFakeProvider) StartStack(string) error          { return nil }
-func (f *volDumpFakeProvider) RefreshAndIsRunning(string) bool  { return true }
+func (f *volDumpFakeProvider) StartStack(string) error         { return nil }
+func (f *volDumpFakeProvider) RefreshAndIsRunning(string) bool { return true }
 func (f *volDumpFakeProvider) GetStackRecoveryInfo(string) (RecoveryInfo, bool) {
 	return RecoveryInfo{}, false
 }
@@ -71,9 +71,9 @@ func TestRunVolumeDumps_GatesPrecedeDump(t *testing.T) {
 		},
 		volumes: map[string][]string{
 			"nextcloud": {"nextcloud_nextcloud_html"},
-			"rallly":    nil,                        // volume-less — must NOT be stopped/dumped
-			"traefik":   {"traefik_data"},           // protected — never considered
-			"diskapp":   {"diskapp_data"},           // volume-bearing but drive disconnected
+			"rallly":    nil,              // volume-less — must NOT be stopped/dumped
+			"traefik":   {"traefik_data"}, // protected — never considered
+			"diskapp":   {"diskapp_data"}, // volume-bearing but drive disconnected
 		},
 		hdd: map[string]string{"nextcloud": usbDrive, "diskapp": badDrive},
 	}
@@ -12,6 +12,9 @@
 {{if .Backup}}{{if .Backup.FlashError}}
 <div class="flash flash-error">{{.Backup.FlashError}}</div>
 {{end}}{{end}}
+{{if .Backup}}{{if .Backup.SingleCopyWarning}}
+<div class="alert alert-warning">{{.Backup.SingleCopyWarning}}</div>
+{{end}}{{end}}

 <!-- Part B: async restore progress banner — polls /api/backup/restore-status; neutral while running,
     red only on failure (exception-color principle). Hidden until a restore op is seen. -->