diff --git a/controller/internal/backup/backup.go b/controller/internal/backup/backup.go
index 9e7e268..e3e1459 100644
--- a/controller/internal/backup/backup.go
+++ b/controller/internal/backup/backup.go
@@ -92,6 +92,15 @@ type Manager struct {
 	// (`-a --delete`, contents-of-src semantics).
 	tier2Mirror func(src, dst string) error
 
+	// sharesPassdbCapture (R-7b) — the samba passdb capture seam (a `docker exec … tar cf -`),
+	// overridable so the shares payload builder is unit-testable without docker. Nil → the real
+	// defaultSharesPassdbCapture. Best-effort by contract: an error yields a manifest-only payload.
+	sharesPassdbCapture func() ([]byte, error)
+
+	// sharesPassdbRestore (R-7b) — the mirror seam for putting a captured passdb archive BACK into the
+	// samba named volume (`docker exec -i … tar xf -`). Nil → the real defaultSharesPassdbRestore.
+	sharesPassdbRestore func(tar []byte) error
+
 	// tier2SSDFits (3b) — the SSD-headroom predicate seam, overridable in tests (system.GetDiskUsage is
 	// Linux-only → nil on the Windows test host, which would always refuse the SSD branch). Nil → the
 	// real tier2FitsSystemDrive.
diff --git a/controller/internal/backup/shares_payload.go b/controller/internal/backup/shares_payload.go
new file mode 100644
index 0000000..4a0cff5
--- /dev/null
+++ b/controller/internal/backup/shares_payload.go
@@ -0,0 +1,265 @@
+package backup
+
+import (
+	"encoding/json"
+	"fmt"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"sort"
+	"strings"
+	"time"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/infra"
+	"gitea.dooplex.hu/admin/felhom-controller/internal/settings"
+)
+
+// SMB share backup — R-7b, Model B′ (Viktor's ruling 2026-07-18). Share data enters the live backup
+// runs through a SIBLING shares source: additive job/leg code that reuses the proven primitives
+// (tier-2 mirror seam, restic wrappers, soft-quota gate, status recording) while leaving every
+// per-app engine path BYTE-IDENTICAL. This file holds the piece both tiers share: the PAYLOAD — the
+// share DEFINITIONS plus the SMB password hash — so a DR rebuild restores the files, the share
+// configuration, and the household credential, not just the bytes on disk.
+//
+// Why a payload at all: the share folders are plain customer directories; copying them protects the
+// data but loses everything that made them shares. Without the manifest a restore leaves the customer
+// with their files back and an empty „Megosztás" page.
+
+const (
+	// SharesPseudoStack is the RESERVED key the shares source occupies wherever the per-app engines
+	// key by stack name: the restic tag, the tier-2 dest root (backups/secondary/_shares/…), and the
+	// CrossDriveBackup status record. It is NEVER shown to the customer — every UI/notification
+	// boundary maps it through SharesDisplayName (see DisplayStackName).
+	//
+	// RESERVED-NAME VERIFICATION (R-7b): the claim that slice-1 validation already excludes a leading
+	// underscore is FALSE for shares — settings.nbNameRe starts with [A-Za-z0-9_], so „_shares" was an
+	// accepted share name. settings.ValidateSMBShareName now rejects the underscore-prefixed namespace
+	// outright. Stack names come from the git-synced catalog (not customer input) and so cannot
+	// realistically produce „_shares", but RunAllTier2 and RunOffboxBackup skip such a stack with a
+	// loud WARN as defense in depth rather than let it clobber the shares tree.
+	SharesPseudoStack = "_shares"
+	// SharesDisplayName is the Hungarian customer-facing name for the shares source.
+	SharesDisplayName = "Megosztások"
+
+	// sharesManifestName is the share-definition document inside the payload.
+	sharesManifestName = "_shares-manifest.json"
+	// sharesPassdbName is the SECRET-BEARING samba passdb archive inside the payload. It rides the
+	// restic repo (encrypted at rest) and the tier-2 staging on the customer's own drives; its bytes
+	// and its name never reach a log line at INFO, a report, or a committed file.
+	sharesPassdbName = "passdb.tar"
+	// sharesPassdbMember is the subtree inside infra.SambaPassdbMount that holds the credential.
+	sharesPassdbMember = "private"
+
+	// sharesManifestVersion pins the payload document shape.
+	sharesManifestVersion = 1
+)
+
+// DisplayStackName maps an engine-internal stack key to the name a customer may see. Today the only
+// mapping is the reserved shares pseudo-stack; every other key is its own display name. THIS is the
+// single boundary that keeps „_shares" out of the Hungarian UI, alerts and e-mails.
+func DisplayStackName(key string) string {
+	if key == SharesPseudoStack {
+		return SharesDisplayName
+	}
+	return key
+}
+
+// SharesManifest is the restorable share-definition document. It is deliberately timestamp-free at
+// the document level so the rendered JSON is BYTE-DETERMINISTIC for an unchanged registry — the
+// tier-2 mirror then has nothing to rewrite on a no-op run. Per-share CreatedAt is preserved.
+type SharesManifest struct {
+	Version    int                 `json:"version"`
+	ServerName string              `json:"server_name"`
+	Shares     []settings.SMBShare `json:"shares"`
+}
+
+// SetSharesPassdbCapturer overrides the samba passdb capture (tests inject a fake so no docker runs).
+func (m *Manager) SetSharesPassdbCapturer(fn func() ([]byte, error)) { m.sharesPassdbCapture = fn }
+
+// sharesPassdbCapturer returns the passdb capture seam (nil → the real docker exec).
+func (m *Manager) sharesPassdbCapturer() func() ([]byte, error) {
+	if m.sharesPassdbCapture != nil {
+		return m.sharesPassdbCapture
+	}
+	return defaultSharesPassdbCapture
+}
+
+// defaultSharesPassdbCapture tars the samba private/ subtree out of the running container on stdout.
+// Best-effort by contract: a stopped container simply yields an error and the payload ships
+// manifest-only (re-setting the SMB password is a cheap UX step; the definitions are the load-bearing
+// part). Uses the same `docker exec` shape as stacks.extractInitialCreds.
+func defaultSharesPassdbCapture() ([]byte, error) {
+	cmd := exec.Command("docker", "exec", infra.SambaContainerName,
+		"tar", "cf", "-", "-C", infra.SambaPassdbMount, sharesPassdbMember)
+	out, err := cmd.Output()
+	if err != nil {
+		return nil, fmt.Errorf("passdb capture: %w", err)
+	}
+	return out, nil
+}
+
+// sharesPayloadDir is the staging directory both tiers read. It lives in the controller DATA DIR
+// (not on a customer drive) for two reasons: the payload is kilobytes — manifest JSON plus a small
+// tdb archive — so the F-A1 rootfs-filler concern does not apply, and a DETERMINISTIC absolute path
+// makes the restic snapshot path stable, which is what the restore mapper anchors on. 0700 because
+// the passdb archive is secret-bearing.
+func (m *Manager) sharesPayloadDir() string {
+	return filepath.Join(m.cfg.Paths.DataDir, "shares-payload")
+}
+
+// SharesPayloadDir exposes the staging path for the restore mapper and tests.
+func (m *Manager) SharesPayloadDir() string { return m.sharesPayloadDir() }
+
+// buildSharesManifest renders the deterministic manifest document from the live registry. Shares are
+// sorted case-insensitively by name so an unchanged registry always marshals to identical bytes
+// regardless of the order the customer happened to add them in.
+func (m *Manager) buildSharesManifest() SharesManifest {
+	mf := SharesManifest{Version: sharesManifestVersion, Shares: []settings.SMBShare{}}
+	if m.settings == nil {
+		return mf
+	}
+	smb := m.settings.GetSMBSettings()
+	mf.ServerName = smb.EffectiveServerName()
+	mf.Shares = append(mf.Shares, m.settings.GetSMBShares()...)
+	sort.Slice(mf.Shares, func(i, j int) bool {
+		a, b := strings.ToLower(mf.Shares[i].Name), strings.ToLower(mf.Shares[j].Name)
+		if a != b {
+			return a < b
+		}
+		return mf.Shares[i].Name < mf.Shares[j].Name
+	})
+	return mf
+}
+
+// buildSharesPayload stages the payload and returns its directory. The manifest is ALWAYS written
+// (it is the definitions-protection floor that must never regress); the passdb archive is
+// best-effort and its absence is a WARN, not an error. Returns the dir plus whether the passdb made
+// it in, so callers can report honestly.
+func (m *Manager) buildSharesPayload() (dir string, passdbOK bool, err error) {
+	dir = m.sharesPayloadDir()
+	if mkErr := os.MkdirAll(dir, 0o700); mkErr != nil {
+		return "", false, fmt.Errorf("shares payload dir: %w", mkErr)
+	}
+	// Tighten an inherited-loose mode from an older layout (the passdb archive lives here).
+	if chErr := os.Chmod(dir, 0o700); chErr != nil {
+		m.logger.Printf("[WARN] [shares] payload dir chmod failed: %v", chErr)
+	}
+
+	blob, mErr := json.MarshalIndent(m.buildSharesManifest(), "", "  ")
+	if mErr != nil {
+		return "", false, fmt.Errorf("shares manifest marshal: %w", mErr)
+	}
+	blob = append(blob, '\n')
+	if wErr := writeFileAtomic(filepath.Join(dir, sharesManifestName), blob, 0o600); wErr != nil {
+		return "", false, fmt.Errorf("shares manifest write: %w", wErr)
+	}
+
+	passdbPath := filepath.Join(dir, sharesPassdbName)
+	tar, pErr := m.sharesPassdbCapturer()()
+	switch {
+	case pErr != nil || len(tar) == 0:
+		// Container down / never deployed. Keep any PREVIOUSLY captured archive rather than deleting
+		// it — a stale credential copy is strictly better for DR than none, and the payload stays
+		// self-consistent because the manifest carries no password of its own.
+		if _, sErr := os.Stat(passdbPath); sErr == nil {
+			m.logger.Printf("[WARN] [shares] passdb capture unavailable (sharing service down?) — keeping the previously captured copy: %v", pErr)
+			passdbOK = true
+		} else {
+			m.logger.Printf("[WARN] [shares] passdb capture unavailable (sharing service down?) — payload is manifest-only; the SMB password must be re-set after a restore: %v", pErr)
+		}
+	default:
+		if wErr := writeFileAtomic(passdbPath, tar, 0o600); wErr != nil {
+			// Never fatal: definitions protection must not hinge on the credential copy.
+			m.logger.Printf("[WARN] [shares] passdb stage failed — payload is manifest-only: %v", wErr)
+		} else {
+			passdbOK = true
+			m.logger.Printf("[DEBUG] [shares] passdb archive staged (%d bytes)", len(tar))
+		}
+	}
+	m.logger.Printf("[INFO] [shares] payload staged: %d share definition(s), credential copy=%v", len(m.buildSharesManifest().Shares), passdbOK)
+	return dir, passdbOK, nil
+}
+
+// writeFileAtomic writes via tmp + rename at the requested mode (mirrors stacks.sambaWriteAtomic's
+// discipline: a crash mid-write can never leave a half-manifest a restore would read).
+func writeFileAtomic(path string, data []byte, mode os.FileMode) error {
+	tmp := path + ".tmp"
+	if err := os.WriteFile(tmp, data, mode); err != nil {
+		return err
+	}
+	if err := os.Chmod(tmp, mode); err != nil {
+		_ = os.Remove(tmp)
+		return err
+	}
+	if err := os.Rename(tmp, path); err != nil {
+		_ = os.Remove(tmp)
+		return err
+	}
+	return nil
+}
+
+// classifiedShare is one share the backup sources act on, with its class resolved from the registry
+// exactly as stacks.sambaClassifiedBinds does it: Offsite=true ⇒ MANDATORY (offsite + tier-2),
+// Offsite=false ⇒ OPTIONAL (tier-2 only). Availability is resolved here too, so BOTH jobs skip a
+// dead mount identically (Scenario F).
+type classifiedShare struct {
+	settings.SMBShare
+	mandatory bool
+}
+
+// classifiedShares returns the shares both jobs act on, dropping unavailable ones with a loud WARN.
+// A share whose owning drive is disconnected/decommissioned, or whose folder simply is not there, is
+// NEVER handed to a mirror or a restic argv: mirroring a missing mountpoint would copy an empty dir
+// over a good backup, and that is the silently-wrong-restore class this codebase refuses by rule.
+func (m *Manager) classifiedShares() []classifiedShare {
+	if m.settings == nil {
+		return nil
+	}
+	var out []classifiedShare
+	for _, sh := range m.settings.GetSMBShares() {
+		if !m.shareBackupAvailable(sh.Path) {
+			m.logger.Printf("[WARN] [shares] share skipped — folder unavailable (drive away or path missing): name=%s", sh.Name)
+			continue
+		}
+		out = append(out, classifiedShare{SMBShare: sh, mandatory: sh.Offsite})
+	}
+	sort.Slice(out, func(i, j int) bool { return strings.ToLower(out[i].Name) < strings.ToLower(out[j].Name) })
+	return out
+}
+
+// shareBackupAvailable mirrors stacks.shareAvailable (the export-time gate) so the backup sources and
+// the smb.conf renderer agree on what "available" means. Kept local rather than imported: the backup
+// package must not depend on the stacks package.
+func (m *Manager) shareBackupAvailable(path string) bool {
+	if strings.TrimSpace(path) == "" {
+		return false
+	}
+	if m.settings != nil {
+		p := filepath.ToSlash(path)
+		for _, sp := range m.settings.GetStoragePaths() {
+			root := filepath.ToSlash(sp.Path)
+			if p == root || strings.HasPrefix(p, root+"/") {
+				if sp.Disconnected || sp.Decommissioned {
+					return false
+				}
+				break
+			}
+		}
+	}
+	st, err := os.Stat(path)
+	return err == nil && st.IsDir()
+}
+
+// sharesEnabled reports whether the shares source has anything to do at all: the feature is on AND at
+// least one share is registered. A disabled feature or an empty registry is a clean no-op in both
+// jobs — no `_shares` restic group, no empty dest dirs (the zero-shares edge case).
+func (m *Manager) sharesEnabled() bool {
+	if m.settings == nil {
+		return false
+	}
+	return m.settings.GetSMBSettings().Enabled && len(m.settings.GetSMBShares()) > 0
+}
+
+// sharesNow is the timestamp helper for the shares status records (kept in one place so tests that
+// assert on recorded status have a single seam to reason about).
+func sharesNow() string { return time.Now().Format(time.RFC3339) }
diff --git a/controller/internal/backup/shares_test.go b/controller/internal/backup/shares_test.go
new file mode 100644
index 0000000..cfb3d8c
--- /dev/null
+++ b/controller/internal/backup/shares_test.go
@@ -0,0 +1,415 @@
+package backup
+
+import (
+	"encoding/json"
+	"io"
+	"log"
+	"os"
+	"path/filepath"
+	"sort"
+	"strings"
+	"testing"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/config"
+	"gitea.dooplex.hu/admin/felhom-controller/internal/settings"
+)
+
+// --- shared harness -------------------------------------------------------------------------------
+
+// sharesEnv is a Manager wired for the R-7b shares source: a real settings store, real temp drives,
+// and FAKE copy/exec seams so nothing shells out to rsync, du, docker or restic.
+type sharesEnv struct {
+	m *Manager
+	// mirrored records every tier2Mirror call as "src=>dst" in call order.
+	mirrored []string
+	// drives maps a label to the temp dir standing in for that registered drive.
+	drives map[string]string
+	sett   *settings.Settings
+	tmp    string
+}
+
+// newSharesEnv registers the named drives, turns sharing on, and installs the fake seams. Shares are
+// added afterwards with addShare so each test states exactly the registry it needs.
+func newSharesEnv(t *testing.T, driveNames ...string) *sharesEnv {
+	t.Helper()
+	tmp := t.TempDir()
+	sett, err := settings.Load(filepath.Join(tmp, "settings.json"), log.New(io.Discard, "", 0))
+	if err != nil {
+		t.Fatal(err)
+	}
+	env := &sharesEnv{drives: map[string]string{}, sett: sett, tmp: tmp}
+	for _, name := range driveNames {
+		p := filepath.Join(tmp, name)
+		if err := os.MkdirAll(p, 0o755); err != nil {
+			t.Fatal(err)
+		}
+		if err := sett.AddStoragePath(settings.StoragePath{Path: p, Label: name, Schedulable: true}); err != nil {
+			t.Fatal(err)
+		}
+		env.drives[name] = p
+	}
+	if err := sett.SetSMBEnabled(true); err != nil {
+		t.Fatal(err)
+	}
+	cfg := &config.Config{}
+	cfg.Paths.DataDir = filepath.Join(tmp, "data")
+	sysDrive := filepath.Join(tmp, "sysdrive")
+	if err := os.MkdirAll(sysDrive, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	cfg.Paths.SystemDataPath = sysDrive
+
+	env.m = &Manager{
+		cfg: cfg, settings: sett, logger: log.New(io.Discard, "", 0), systemDataPath: sysDrive,
+		// Fake mirror: records the call AND actually copies, so restore tests have real bytes.
+		tier2Mirror: func(src, dst string) error {
+			env.mirrored = append(env.mirrored, src+"=>"+dst)
+			return copyTreeForTest(src, dst)
+		},
+		sharesPassdbCapture: func() ([]byte, error) { return []byte("FAKE-PASSDB-TAR"), nil },
+	}
+	return env
+}
+
+// addShare registers a share, creating its folder with one file so it is "available".
+func (e *sharesEnv) addShare(t *testing.T, drive, name string, offsite bool) string {
+	t.Helper()
+	p := filepath.Join(e.drives[drive], name)
+	if err := os.MkdirAll(p, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(p, name+".txt"), []byte("content-of-"+name), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if err := e.sett.AddSMBShare(settings.SMBShare{Name: name, Path: p, Offsite: offsite, CreatedAt: "2026-07-18T00:00:00Z"}); err != nil {
+		t.Fatal(err)
+	}
+	return p
+}
+
+// copyTreeForTest is a minimal recursive copy standing in for rsync in the mirror seam.
+func copyTreeForTest(src, dst string) error {
+	fi, err := os.Stat(src)
+	if err != nil {
+		return err
+	}
+	if !fi.IsDir() {
+		b, rErr := os.ReadFile(src)
+		if rErr != nil {
+			return rErr
+		}
+		if mErr := os.MkdirAll(filepath.Dir(dst), 0o755); mErr != nil {
+			return mErr
+		}
+		return os.WriteFile(dst, b, 0o644)
+	}
+	if err := os.MkdirAll(dst, 0o755); err != nil {
+		return err
+	}
+	entries, err := os.ReadDir(src)
+	if err != nil {
+		return err
+	}
+	for _, e := range entries {
+		if err := copyTreeForTest(filepath.Join(src, e.Name()), filepath.Join(dst, e.Name())); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+// mirroredDsts returns just the destination side of every recorded mirror call.
+func (e *sharesEnv) mirroredDsts() []string {
+	var out []string
+	for _, s := range e.mirrored {
+		out = append(out, s[strings.Index(s, "=>")+2:])
+	}
+	return out
+}
+
+// --- Part 1: payload ------------------------------------------------------------------------------
+
+// The manifest must carry the share DEFINITIONS verbatim — that is the whole point of the payload:
+// after a DR rebuild the customer gets their files AND their „Megosztás" page back, not an empty one.
+func TestSharesPayloadManifestMatchesRegistry(t *testing.T) {
+	env := newSharesEnv(t, "hdd_1")
+	env.addShare(t, "hdd_1", "dokumentumok", true)
+	env.addShare(t, "hdd_1", "filmek", false)
+
+	dir, passdbOK, err := env.m.buildSharesPayload()
+	if err != nil {
+		t.Fatalf("buildSharesPayload: %v", err)
+	}
+	if !passdbOK {
+		t.Fatal("expected the (faked) passdb capture to succeed")
+	}
+	blob, err := os.ReadFile(filepath.Join(dir, sharesManifestName))
+	if err != nil {
+		t.Fatalf("read manifest: %v", err)
+	}
+	var mf SharesManifest
+	if err := json.Unmarshal(blob, &mf); err != nil {
+		t.Fatalf("manifest is not valid JSON: %v", err)
+	}
+	if mf.Version != sharesManifestVersion {
+		t.Errorf("manifest version = %d, want %d", mf.Version, sharesManifestVersion)
+	}
+	if mf.ServerName != settings.DefaultSMBServerName {
+		t.Errorf("server name = %q, want %q", mf.ServerName, settings.DefaultSMBServerName)
+	}
+	want := env.sett.GetSMBShares()
+	sort.Slice(want, func(i, j int) bool { return want[i].Name < want[j].Name })
+	if len(mf.Shares) != len(want) {
+		t.Fatalf("manifest has %d shares, registry has %d", len(mf.Shares), len(want))
+	}
+	for i := range want {
+		if mf.Shares[i] != want[i] {
+			t.Errorf("share %d: manifest %+v != registry %+v", i, mf.Shares[i], want[i])
+		}
+	}
+	// The credential copy must be present and 0600 — it is secret-bearing.
+	st, err := os.Stat(filepath.Join(dir, sharesPassdbName))
+	if err != nil {
+		t.Fatalf("passdb archive missing: %v", err)
+	}
+	if runtimeIsPOSIX() && st.Mode().Perm() != 0o600 {
+		t.Errorf("passdb archive mode = %o, want 600 (secret-bearing)", st.Mode().Perm())
+	}
+}
+
+// Determinism: an unchanged registry must marshal to IDENTICAL bytes regardless of the order shares
+// were added in, so a no-op nightly run gives the tier-2 mirror nothing to rewrite.
+func TestSharesPayloadManifestIsDeterministic(t *testing.T) {
+	// Built from the registry only (no filesystem), so the two runs differ ONLY in insertion order —
+	// otherwise the temp-dir paths baked into each share would make any comparison meaningless.
+	read := func(order []string) []byte {
+		sett, err := settings.Load(filepath.Join(t.TempDir(), "settings.json"), log.New(io.Discard, "", 0))
+		if err != nil {
+			t.Fatal(err)
+		}
+		for _, n := range order {
+			if err := sett.AddSMBShare(settings.SMBShare{
+				Name: n, Path: "/mnt/hdd_1/" + n, Offsite: true, CreatedAt: "2026-07-18T00:00:00Z",
+			}); err != nil {
+				t.Fatal(err)
+			}
+		}
+		m := &Manager{settings: sett, logger: log.New(io.Discard, "", 0)}
+		b, err := json.MarshalIndent(m.buildSharesManifest(), "", "  ")
+		if err != nil {
+			t.Fatal(err)
+		}
+		return b
+	}
+	a := read([]string{"alpha", "zulu", "mike"})
+	b := read([]string{"zulu", "mike", "alpha"})
+	if string(a) != string(b) {
+		t.Errorf("manifest is order-dependent:\n%s\n---\n%s", a, b)
+	}
+}
+
+// Scenario: samba container down. The payload must still be produced (manifest-only) and must NOT
+// error — re-setting the SMB password is cheap; losing the definitions is not.
+func TestSharesPayloadPassdbAbsentIsManifestOnly(t *testing.T) {
+	env := newSharesEnv(t, "hdd_1")
+	env.addShare(t, "hdd_1", "dokumentumok", true)
+	env.m.sharesPassdbCapture = func() ([]byte, error) { return nil, os.ErrNotExist }
+
+	dir, passdbOK, err := env.m.buildSharesPayload()
+	if err != nil {
+		t.Fatalf("a down container must not fail the payload: %v", err)
+	}
+	if passdbOK {
+		t.Error("passdbOK must be false when the capture failed and no prior copy exists")
+	}
+	if _, err := os.Stat(filepath.Join(dir, sharesManifestName)); err != nil {
+		t.Errorf("manifest must still be written: %v", err)
+	}
+	if _, err := os.Stat(filepath.Join(dir, sharesPassdbName)); !os.IsNotExist(err) {
+		t.Error("no passdb archive should exist when the capture failed and none was staged before")
+	}
+}
+
+// A previously captured credential copy must SURVIVE a later failed capture — a stale credential is
+// strictly better for DR than none, and the manifest carries no password of its own to contradict it.
+func TestSharesPayloadKeepsPriorPassdbOnCaptureFailure(t *testing.T) {
+	env := newSharesEnv(t, "hdd_1")
+	env.addShare(t, "hdd_1", "dokumentumok", true)
+	dir, _, err := env.m.buildSharesPayload()
+	if err != nil {
+		t.Fatal(err)
+	}
+	env.m.sharesPassdbCapture = func() ([]byte, error) { return nil, os.ErrNotExist }
+	_, passdbOK, err := env.m.buildSharesPayload()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !passdbOK {
+		t.Error("a previously staged passdb copy must keep passdbOK true")
+	}
+	b, err := os.ReadFile(filepath.Join(dir, sharesPassdbName))
+	if err != nil || string(b) != "FAKE-PASSDB-TAR" {
+		t.Errorf("prior passdb copy was lost: %q, %v", b, err)
+	}
+}
+
+// The reserved key must never reach a Hungarian surface raw.
+func TestDisplayStackNameMapsReservedKey(t *testing.T) {
+	if got := DisplayStackName(SharesPseudoStack); got != SharesDisplayName {
+		t.Errorf("DisplayStackName(%q) = %q, want %q", SharesPseudoStack, got, SharesDisplayName)
+	}
+	if got := DisplayStackName("immich"); got != "immich" {
+		t.Errorf("DisplayStackName must be identity for ordinary stacks, got %q", got)
+	}
+}
+
+// --- Part 2: tier-2 shares job --------------------------------------------------------------------
+
+// Scenario B: an OPTIONAL (Felhőmentés-off) share is tier-2'd like any other — the tier-2 tier is the
+// local copy every share gets; only the offsite tier discriminates.
+func TestSharesTier2MirrorsBothClasses(t *testing.T) {
+	env := newSharesEnv(t, "hdd_1", "hdd_2")
+	env.addShare(t, "hdd_1", "dokumentumok", true)
+	env.addShare(t, "hdd_1", "filmek", false)
+
+	if err := env.m.RunSharesTier2(); err != nil {
+		t.Fatalf("RunSharesTier2: %v", err)
+	}
+	key := sharesDriveKey(env.drives["hdd_1"])
+	destBase := filepath.Join(NamespaceRoot(env.drives["hdd_2"], true), "backups", "secondary", SharesPseudoStack)
+
+	for _, share := range []string{"dokumentumok", "filmek"} {
+		want := filepath.Join(destBase, key, share)
+		if _, err := os.Stat(filepath.Join(want, share+".txt")); err != nil {
+			t.Errorf("share %s was not mirrored to %s: %v", share, want, err)
+		}
+	}
+	// The payload rides the target so it is independently restorable.
+	if _, err := os.Stat(filepath.Join(destBase, sharesPayloadDestRel, sharesManifestName)); err != nil {
+		t.Errorf("payload manifest missing from the tier-2 target: %v", err)
+	}
+	// Marker LAST: its presence means every leg + reconcile completed.
+	b, err := os.ReadFile(filepath.Join(destBase, tier2LayoutMarker))
+	if err != nil || string(b) != tier2LayoutVersion {
+		t.Errorf("layout marker missing/wrong: %q, %v", b, err)
+	}
+	// Status is recorded under the reserved key, and renders as „Megosztások".
+	cd := env.m.SharesTier2Status()
+	if cd == nil || cd.LastStatus != "ok" {
+		t.Fatalf("shares tier-2 status = %+v, want LastStatus=ok", cd)
+	}
+}
+
+// The target may NEVER be the leg's own source drive — that would be a same-disk copy pretending to
+// be a second backup. With shares on two drives each group must land on the other one.
+func TestSharesTier2NeverTargetsItsOwnSourceDrive(t *testing.T) {
+	env := newSharesEnv(t, "hdd_1", "hdd_2")
+	env.addShare(t, "hdd_1", "egy", true)
+	env.addShare(t, "hdd_2", "ketto", true)
+
+	if err := env.m.RunSharesTier2(); err != nil {
+		t.Fatalf("RunSharesTier2: %v", err)
+	}
+	for _, call := range env.mirrored {
+		src, dst, _ := strings.Cut(call, "=>")
+		if strings.HasPrefix(src, env.drives["hdd_1"]) && strings.HasPrefix(dst, env.drives["hdd_1"]) {
+			t.Errorf("leg mirrored onto its own source drive: %s", call)
+		}
+		if strings.HasPrefix(src, env.drives["hdd_2"]) && strings.HasPrefix(dst, env.drives["hdd_2"]) {
+			t.Errorf("leg mirrored onto its own source drive: %s", call)
+		}
+	}
+}
+
+// Scenario F: a share whose drive is marked away is skipped LOUDLY and its path reaches NO mirror
+// argument — mirroring a dead mountpoint would copy an empty dir over a good backup.
+func TestSharesTier2SkipsDeadMountAndContinues(t *testing.T) {
+	env := newSharesEnv(t, "hdd_1", "hdd_2", "hdd_3")
+	env.addShare(t, "hdd_1", "elo", true)
+	deadPath := env.addShare(t, "hdd_3", "halott", true)
+	if err := env.sett.SetDisconnected(env.drives["hdd_3"], true, nil); err != nil {
+		t.Fatal(err)
+	}
+
+	if err := env.m.RunSharesTier2(); err != nil {
+		t.Fatalf("the run must continue for healthy shares: %v", err)
+	}
+	for _, call := range env.mirrored {
+		if strings.Contains(call, deadPath) {
+			t.Errorf("dead-mount share reached a mirror call: %s", call)
+		}
+	}
+	var sawLive bool
+	for _, dst := range env.mirroredDsts() {
+		if strings.HasSuffix(dst, string(filepath.Separator)+"elo") {
+			sawLive = true
+		}
+	}
+	if !sawLive {
+		t.Error("the healthy share must still be mirrored")
+	}
+}
+
+// Zero shares / sharing disabled: a clean no-op — no dest tree, no status record, no empty dirs.
+func TestSharesTier2ZeroSharesIsCleanNoOp(t *testing.T) {
+	env := newSharesEnv(t, "hdd_1", "hdd_2")
+	if err := env.m.RunSharesTier2(); err != nil {
+		t.Fatalf("RunSharesTier2: %v", err)
+	}
+	if len(env.mirrored) != 0 {
+		t.Errorf("expected no mirror calls, got %v", env.mirrored)
+	}
+	if cd := env.m.SharesTier2Status(); cd != nil {
+		t.Errorf("expected no status record for a feature nobody uses, got %+v", cd)
+	}
+	destBase := filepath.Join(NamespaceRoot(env.drives["hdd_2"], true), "backups", "secondary", SharesPseudoStack)
+	if _, err := os.Stat(destBase); !os.IsNotExist(err) {
+		t.Errorf("expected no _shares dest tree, stat err = %v", err)
+	}
+}
+
+// A share removed from the registry must stop occupying the secondary drive within ONE run — the
+// reconcile pass prunes it, reusing the same destBase-bounded removal guard the app path uses.
+func TestSharesTier2ReconcilePrunesRemovedShare(t *testing.T) {
+	env := newSharesEnv(t, "hdd_1", "hdd_2")
+	env.addShare(t, "hdd_1", "marad", true)
+	env.addShare(t, "hdd_1", "torolt", true)
+	if err := env.m.RunSharesTier2(); err != nil {
+		t.Fatal(err)
+	}
+	key := sharesDriveKey(env.drives["hdd_1"])
+	destBase := filepath.Join(NamespaceRoot(env.drives["hdd_2"], true), "backups", "secondary", SharesPseudoStack)
+	stale := filepath.Join(destBase, key, "torolt")
+	if _, err := os.Stat(stale); err != nil {
+		t.Fatalf("precondition: %s should exist after the first run: %v", stale, err)
+	}
+
+	if err := env.sett.RemoveSMBShare("torolt"); err != nil {
+		t.Fatal(err)
+	}
+	if err := env.m.RunSharesTier2(); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := os.Stat(stale); !os.IsNotExist(err) {
+		t.Errorf("removed share's dest dir was not pruned: stat err = %v", err)
+	}
+	if _, err := os.Stat(filepath.Join(destBase, key, "marad")); err != nil {
+		t.Errorf("surviving share was pruned by mistake: %v", err)
+	}
+}
+
+// Drive keys must be collision-free across drives that share a basename.
+func TestSharesDriveKeyIsCollisionFree(t *testing.T) {
+	if sharesDriveKey("/mnt/a/data") == sharesDriveKey("/mnt/b/data") {
+		t.Error("drives with the same basename must not map to the same dest key")
+	}
+	for _, bad := range []string{"/", "\x00", " "} {
+		if k := sharesDriveKey(bad); strings.ContainsAny(k, `/\`) {
+			t.Errorf("sharesDriveKey(%q) = %q — must be a single safe path segment", bad, k)
+		}
+	}
+}
+
+// runtimeIsPOSIX reports whether file-mode assertions are meaningful on this host (Windows reports
+// synthesised permissions, so mode checks there are noise rather than signal).
+func runtimeIsPOSIX() bool { return os.PathSeparator == '/' }
diff --git a/controller/internal/backup/tier2.go b/controller/internal/backup/tier2.go
index f903441..2646dd3 100644
--- a/controller/internal/backup/tier2.go
+++ b/controller/internal/backup/tier2.go
@@ -86,6 +86,17 @@ func tier2FitsHeadroom(availGB, totalGB, unitGB float64) bool {
 // silently-wrong-owner restore). fullSize sizes the real-drive path; stateOnlySize sizes the SSD path.
 func (m *Manager) selectTier2Target(stackName string, fullSize, stateOnlySize int64) (*Tier2Target, error) {
 	sourceDrive := m.GetAppDrivePath(stackName)
+	if sourceDrive == "" {
+		return nil, fmt.Errorf("no source drive for %s", stackName)
+	}
+	return m.selectTier2TargetFrom(stackName, sourceDrive, fullSize, stateOnlySize)
+}
+
+// selectTier2TargetFrom is selectTier2Target with the source drive supplied EXPLICITLY. It exists so
+// the R-7b shares source — whose "source drive" is the drive a group of shares lives on, not an app's
+// GetAppDrivePath — can reuse this selection and its headroom math verbatim instead of forking it.
+// The app path above is a thin wrapper; nothing about its behaviour changed.
+func (m *Manager) selectTier2TargetFrom(stackName, sourceDrive string, fullSize, stateOnlySize int64) (*Tier2Target, error) {
 	if sourceDrive == "" {
 		return nil, fmt.Errorf("no source drive for %s", stackName)
 	}
@@ -232,6 +243,13 @@ func classifyTier2Rel(dirRel string, legRels []string) tier2RelClass {
 // descendant of any current leg relpath (§7-D — the deferred-pruning answer: a bind removed/re-classed
 // stops occupying the secondary drive within one run). Runs strictly inside destBase.
 func (m *Manager) tier2Reconcile(destBase string, legRels []string) {
+	m.tier2ReconcileRoots(destBase, []string{"hdd", "userdata"}, legRels)
+}
+
+// tier2ReconcileRoots is tier2Reconcile with the top-level dest roots supplied explicitly — a pure
+// extraction so the R-7b shares dest (whose roots are per-source-drive keys, not hdd/userdata) can
+// reuse the SAME staleness classification and the SAME destBase-bounded removal guard.
+func (m *Manager) tier2ReconcileRoots(destBase string, roots, legRels []string) {
 	var walk func(dirAbs, dirRel string)
 	walk = func(dirAbs, dirRel string) {
 		entries, err := os.ReadDir(dirAbs)
@@ -258,7 +276,7 @@ func (m *Manager) tier2Reconcile(destBase string, legRels []string) {
 			}
 		}
 	}
-	for _, root := range []string{"hdd", "userdata"} {
+	for _, root := range roots {
 		walk(filepath.Join(destBase, root), root)
 	}
 }
@@ -397,6 +415,14 @@ func (m *Manager) RunAllTier2() {
 	}
 	var n int
 	for _, stack := range m.stackProvider.ListDeployedStacks() {
+		// Reserved-name defense in depth (R-7b): the shares source owns backups/secondary/_shares and
+		// the _shares status record. Stack names come from the git-synced catalog, not customer input,
+		// so this cannot realistically fire — but if it ever did, the app would silently overwrite the
+		// shares tree, so it is refused loudly instead.
+		if stack.Name == SharesPseudoStack {
+			m.logger.Printf("[ERROR] [backup] Tier 2: stack %q uses the RESERVED shares key — skipped to protect the shares backup tree", stack.Name)
+			continue
+		}
 		// 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
@@ -416,6 +442,12 @@ func (m *Manager) RunAllTier2() {
 		n++
 	}
 	m.logger.Printf("[INFO] [backup] Tier 2 run complete: %d app(s) processed (incl. volume-only — F6)", n)
+
+	// R-7b: the SHARES source runs after the per-app loop, in the SAME orchestrator run. It is a
+	// sibling job — nothing above it changed — and its failure never fails the app tier.
+	if err := m.RunSharesTier2(); err != nil {
+		m.logger.Printf("[WARN] [backup] Tier 2 shares job failed: %v", err)
+	}
 }
 
 // --- per-app config-panel view (drives the Tier-2 "Beállítás" page) ---
diff --git a/controller/internal/backup/tier2_shares.go b/controller/internal/backup/tier2_shares.go
new file mode 100644
index 0000000..b9794b5
--- /dev/null
+++ b/controller/internal/backup/tier2_shares.go
@@ -0,0 +1,291 @@
+package backup
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"sort"
+	"strings"
+	"time"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/settings"
+	"gitea.dooplex.hu/admin/felhom-controller/internal/system"
+)
+
+// Tier-2 shares job — R-7b Part 2, the local cross-drive leg of Model B′.
+//
+// It runs AFTER the per-stack tier-2 loop, inside the SAME orchestrator run, and it is a SIBLING of
+// that loop, never a modification of it: RunTier2 and its per-app dest tree are untouched (the B′
+// invariant, enforced by TestSharesTier2LeavesPerAppTreeUntouched). What it reuses instead is every
+// primitive the per-app path proved: the tier2Mirror seam, selectTier2TargetFrom's target choice and
+// headroom math, tier2ReconcileRoots' staleness pruning, tier2SafeRemove's destBase-bounded removal,
+// the marker-written-LAST discipline, and the recordTier2* status helpers.
+//
+// Layout — backups/secondary/_shares/ on the target drive:
+//
+//	.felhom-tier2-layout        marker, content "2", written LAST (after every leg + reconcile)
+//	_payload/                   the share definitions + credential copy (shares_payload.go)
+//	<sourceDriveKey>/<share>/   one mirrored leg per share, grouped by the drive it came from
+//
+// Shares are grouped BY SOURCE DRIVE because a household's shares can span several drives and each
+// group needs its own cross-drive target: a leg's target may never be that leg's own source drive.
+
+// sharesPayloadDestRel is the payload's relpath inside the shares dest (a reserved root name; the
+// leading underscore cannot collide with a drive key because drive keys are derived from paths).
+const sharesPayloadDestRel = "_payload"
+
+// sharesDriveKey turns an absolute source-drive path into ONE safe dest path segment. The full path
+// is encoded (not just its basename) so two drives whose mountpoints share a basename — /mnt/a/data
+// and /mnt/b/data — can never map onto the same dest subtree and silently overwrite each other.
+// Deterministic and stable across runs, which is what lets the reconcile pass recognise its own dirs.
+func sharesDriveKey(drive string) string {
+	s := strings.Trim(filepath.ToSlash(drive), "/")
+	if s == "" {
+		return "root"
+	}
+	var b strings.Builder
+	for _, r := range s {
+		switch {
+		case r >= 'a' && r <= 'z', r >= 'A' && r <= 'Z', r >= '0' && r <= '9', r == '.', r == '-':
+			b.WriteRune(r)
+		default:
+			b.WriteByte('_')
+		}
+	}
+	return b.String()
+}
+
+// shareSourceDrive resolves the registered storage root a share's folder lives under. Returns "" when
+// the share sits under no registered root — such a share has no meaningful "other drive" and is
+// skipped with a warning rather than guessed at.
+func (m *Manager) shareSourceDrive(path string) string {
+	if m.settings == nil {
+		return ""
+	}
+	p := filepath.ToSlash(path)
+	best := ""
+	for _, sp := range m.settings.GetStoragePaths() {
+		root := filepath.ToSlash(sp.Path)
+		if root == "" {
+			continue
+		}
+		// Longest match wins, so a nested registered root claims its own shares. Containment is
+		// compared in slash form, but the returned value is the REGISTRY'S OWN string: the target
+		// selector compares the source drive against sp.Path by equality, and handing it a normalised
+		// variant would make that comparison miss and let a group target its own source drive.
+		if (p == root || strings.HasPrefix(p, root+"/")) && len(root) > len(best) {
+			best = sp.Path
+		}
+	}
+	return best
+}
+
+// sharesTier2Group is one source drive's worth of shares plus the target chosen for it.
+type sharesTier2Group struct {
+	sourceDrive string
+	key         string
+	shares      []classifiedShare
+}
+
+// RunSharesTier2 mirrors every classified, available share to a cross-drive target and stages the
+// payload beside it. Best-effort and idempotent, exactly like RunTier2: an absent target is an
+// honest recorded status, not an error. Returns the first hard copy error.
+func (m *Manager) RunSharesTier2() error {
+	if !m.sharesEnabled() {
+		// Feature off or no shares registered: a clean no-op. Deliberately NOT a recorded status —
+		// writing one would make the „Megosztás" page claim a backup tier for a feature in use by
+		// nobody, and would create the dest tree for zero shares.
+		return nil
+	}
+	if m.settings != nil {
+		if cd := m.settings.GetCrossDriveConfig(SharesPseudoStack); cd != nil && cd.UserDisabled {
+			m.logger.Printf("[INFO] [shares] tier-2 skipped — disabled by customer")
+			return nil
+		}
+	}
+	shares := m.classifiedShares()
+	if len(shares) == 0 {
+		// Every registered share is on a drive that is away / a folder that vanished. That is a real
+		// operational state the customer must see, not silence.
+		m.recordTier2NoTarget(SharesPseudoStack, "egyetlen megosztás mappája sem érhető el — ellenőrizd a meghajtókat")
+		m.logger.Printf("[WARN] [shares] tier-2: no available share folders — nothing mirrored")
+		return nil
+	}
+
+	// Group by source drive (deterministic order so logs and dest trees are reproducible).
+	byDrive := map[string]*sharesTier2Group{}
+	var warns []string
+	for _, sh := range shares {
+		drive := m.shareSourceDrive(sh.Path)
+		if drive == "" {
+			m.logger.Printf("[WARN] [shares] tier-2: share %s is not under a registered storage root — skipped", sh.Name)
+			warns = append(warns, fmt.Sprintf("A(z) „%s” megosztás nem regisztrált adatmeghajtón van — kimaradt a 2. mentésből.", sh.Name))
+			continue
+		}
+		g := byDrive[drive]
+		if g == nil {
+			g = &sharesTier2Group{sourceDrive: drive, key: sharesDriveKey(drive)}
+			byDrive[drive] = g
+		}
+		g.shares = append(g.shares, sh)
+	}
+	var groups []*sharesTier2Group
+	for _, g := range byDrive {
+		groups = append(groups, g)
+	}
+	sort.Slice(groups, func(i, j int) bool { return groups[i].sourceDrive < groups[j].sourceDrive })
+	if len(groups) == 0 {
+		m.recordTier2NoTarget(SharesPseudoStack, "a megosztások egyike sincs regisztrált adatmeghajtón")
+		return nil
+	}
+
+	// Payload once per run — both tiers read the same staged directory.
+	payloadDir, _, perr := m.buildSharesPayload()
+	if perr != nil {
+		m.logger.Printf("[WARN] [shares] tier-2: payload staging failed — mirroring files without the definition manifest: %v", perr)
+		warns = append(warns, "A megosztás-beállítások mentése nem sikerült — a fájlok mentése megtörtént.")
+		payloadDir = ""
+	}
+
+	mirror := m.tier2Mirror
+	if mirror == nil {
+		mirror = rsyncMirror
+	}
+	start := time.Now()
+	var (
+		totalSize   int64
+		lastTarget  *Tier2Target
+		noTargetWhy []string
+		mirrored    int
+	)
+	for _, g := range groups {
+		// Two sizes, mirroring RunTier2's contract: full = every share in the group; state-only = the
+		// MANDATORY (Felhőmentés-on) subset, which is what the SSD headroom guard must fit.
+		var fullSize, stateOnlySize int64
+		for _, sh := range g.shares {
+			sz := dirSizeBytes(sh.Path)
+			fullSize += sz
+			if sh.mandatory {
+				stateOnlySize += sz
+			}
+		}
+		target, err := m.selectTier2TargetFrom(SharesPseudoStack, g.sourceDrive, fullSize, stateOnlySize)
+		if err != nil {
+			why := tier2NoTargetReason(err)
+			noTargetWhy = append(noTargetWhy, fmt.Sprintf("%s: %s", g.sourceDrive, why))
+			m.logger.Printf("[INFO] [shares] tier-2: no off-drive target for shares on %s — %s", g.sourceDrive, why)
+			continue
+		}
+		// Defense-in-depth off-drive guard (selection already enforced it): a leg's target may never
+		// be that leg's own source drive — that would be a same-disk "copy" pretending to be tier 2.
+		if system.SamePhysicalDevice(g.sourceDrive, target.NamespaceRoot) {
+			noTargetWhy = append(noTargetWhy, fmt.Sprintf("%s: a kiválasztott cél ugyanazon a fizikai lemezen van", g.sourceDrive))
+			continue
+		}
+		legs := g.shares
+		if target.StateOnly {
+			kept := legs[:0]
+			dropped := false
+			for _, sh := range legs {
+				if sh.mandatory {
+					kept = append(kept, sh)
+				} else {
+					dropped = true
+				}
+			}
+			legs = kept
+			if dropped {
+				warns = append(warns, "A belső SSD-re csak a felhőmentésre jelölt megosztások férnek el — a többi nem került másolásra.")
+			}
+		}
+
+		destBase := filepath.Join(target.NamespaceRoot, "backups", "secondary", SharesPseudoStack)
+		legRels := make([]string, 0, len(legs)+1)
+		for _, sh := range legs {
+			rel := g.key + "/" + sh.Name
+			if err := mirror(sh.Path, filepath.Join(destBase, filepath.FromSlash(rel))); err != nil {
+				m.recordTier2Failure(SharesPseudoStack, target, err)
+				if m.tier2Notify != nil {
+					m.tier2Notify(SharesPseudoStack, target.Label, time.Since(start), err)
+				}
+				return fmt.Errorf("tier2 shares mirror %s: %w", sh.Name, err)
+			}
+			legRels = append(legRels, rel)
+			totalSize += dirSizeBytes(sh.Path)
+			mirrored++
+		}
+		// The payload rides every target so each one is independently restorable.
+		if payloadDir != "" {
+			if err := mirror(payloadDir, filepath.Join(destBase, sharesPayloadDestRel)); err != nil {
+				m.logger.Printf("[WARN] [shares] tier-2: payload mirror to %s failed (files are copied): %v", destBase, err)
+			} else {
+				legRels = append(legRels, sharesPayloadDestRel)
+			}
+		}
+		// Prune dest dirs no share covers any more (a share deleted or renamed since the last run
+		// stops occupying the secondary drive within one run), then the marker LAST — a half-written
+		// dest therefore has no marker and the restore path refuses it until the next good run.
+		m.tier2ReconcileRoots(destBase, sharesTier2DestRoots(destBase), legRels)
+		if err := m.writeTier2Marker(destBase); err != nil {
+			m.logger.Printf("[WARN] [shares] tier-2: layout marker write failed (restore will refuse until next run): %v", err)
+		}
+		lastTarget = target
+		m.logger.Printf("[INFO] [shares] tier-2 copied %d share(s) from %s → %s (%s)",
+			len(legs), g.sourceDrive, destBase, humanizeBytes(totalSize))
+	}
+
+	dur := time.Since(start)
+	if lastTarget == nil {
+		reason := strings.Join(noTargetWhy, "; ")
+		if reason == "" {
+			reason = "nincs másik fizikai meghajtó — a 2. mentéshez 2. meghajtó szükséges"
+		}
+		m.recordTier2NoTarget(SharesPseudoStack, reason)
+		m.logger.Printf("[INFO] [shares] tier-2: no off-drive target for any share group — %s", reason)
+		return nil
+	}
+	if len(noTargetWhy) > 0 {
+		warns = append(warns, "Néhány meghajtón lévő megosztásnak nincs másodlagos célja: "+strings.Join(noTargetWhy, "; "))
+	}
+	m.recordTier2Success(SharesPseudoStack, lastTarget, totalSize, strings.Join(warns, " "), dur)
+	if m.tier2Notify != nil {
+		m.tier2Notify(SharesPseudoStack, lastTarget.Label, dur, nil)
+	}
+	m.logger.Printf("[INFO] [shares] tier-2 run complete: %d share leg(s), %s, %s",
+		mirrored, humanizeBytes(totalSize), dur.Round(time.Second))
+	return nil
+}
+
+// sharesTier2DestRoots lists the top-level dirs under a shares destBase that the reconcile pass may
+// walk. It reads what is ON DISK rather than what this run produced, so a drive key from a drive that
+// is no longer registered still gets visited (and pruned) instead of lingering forever.
+func sharesTier2DestRoots(destBase string) []string {
+	entries, err := os.ReadDir(destBase)
+	if err != nil {
+		return nil
+	}
+	var roots []string
+	for _, e := range entries {
+		if e.IsDir() {
+			roots = append(roots, e.Name())
+		}
+	}
+	sort.Strings(roots)
+	return roots
+}
+
+// writeTier2Marker writes the shared layout marker (content "2") — the LAST write of a dest, so its
+// presence means "every leg and the reconcile completed".
+func (m *Manager) writeTier2Marker(destBase string) error {
+	return os.WriteFile(filepath.Join(destBase, tier2LayoutMarker), []byte(tier2LayoutVersion), 0o644)
+}
+
+// SharesTier2Status returns the recorded shares tier-2 status for the „Megosztás" page (nil when the
+// job has never run). It is the SAME CrossDriveBackup record the per-app rows use, keyed by the
+// reserved pseudo-stack.
+func (m *Manager) SharesTier2Status() *settings.CrossDriveBackup {
+	if m.settings == nil {
+		return nil
+	}
+	return m.settings.GetCrossDriveConfig(SharesPseudoStack)
+}
diff --git a/controller/internal/infra/samba.go b/controller/internal/infra/samba.go
index b74b7de..2b01177 100644
--- a/controller/internal/infra/samba.go
+++ b/controller/internal/infra/samba.go
@@ -30,6 +30,21 @@ type SambaData struct {
 // SambaHouseholdUser is the single household SMB account name (matches the entrypoint's unix user).
 const SambaHouseholdUser = "felhom"
 
+// SambaContainerName is the fixed container name the compose render pins. It is THE single source of
+// truth for it: the renderer below interpolates this constant, stacks.sambaContainer aliases it for
+// the exec paths, monitor's effective-protected set watches it (R-7b liveness), and the backup
+// package execs it for the passdb capture/restore. Anything that needs "which container is samba"
+// reads this — never a second string literal.
+const SambaContainerName = "felhom-samba"
+
+// SambaPassdbVolume / SambaPassdbMount name the docker volume that holds the household SMB credential
+// (the passdb tdb set) and where the container mounts it. R-7b's payload capture/restore targets this
+// mount point, so it is pinned here beside the render that creates it rather than duplicated there.
+const (
+	SambaPassdbVolume = "samba-passdb"
+	SambaPassdbMount  = "/var/lib/samba"
+)
+
 // RenderSambaConfig renders smb.conf: the hardened global block (bind interfaces only = lo eth0,
 // SMB2+ floor, NetBIOS on for flat-name resolution) plus one [section] per share. Deterministic:
 // shares are emitted in the given order (the caller preserves registry order). force user/group pin
@@ -95,21 +110,21 @@ func RenderSambaCompose(d SambaData) string {
 # receive the LAN multicast that WSD/mDNS discovery needs.
 
 services:
-  felhom-samba:
-    image: %s
-    container_name: felhom-samba
+  %[1]s:
+    image: %[2]s
+    container_name: %[1]s
     restart: unless-stopped
     network_mode: host
     environment:
-      - FELHOM_SERVER_NAME=%s
+      - FELHOM_SERVER_NAME=%[3]s
       - FELHOM_IFACE=eth0
-      - FELHOM_UID=%d
-      - FELHOM_GID=%d
+      - FELHOM_UID=%[4]d
+      - FELHOM_GID=%[4]d
     volumes:
       - ./smb.conf:/etc/samba/smb.conf:ro
-      - samba-passdb:/var/lib/samba
-%s
+      - %[5]s:%[6]s
+%[7]s
 volumes:
-  samba-passdb:
-`, SambaImage, d.ServerName, d.UID, d.UID, binds.String())
+  %[5]s:
+`, SambaContainerName, SambaImage, d.ServerName, d.UID, SambaPassdbVolume, SambaPassdbMount, binds.String())
 }
diff --git a/controller/internal/settings/smb.go b/controller/internal/settings/smb.go
index e0d43b9..6306a9c 100644
--- a/controller/internal/settings/smb.go
+++ b/controller/internal/settings/smb.go
@@ -66,6 +66,15 @@ func ValidateSMBShareName(name string) error {
 	if strings.ContainsAny(name, `/\.` ) {
 		return fmt.Errorf("a megosztás neve nem tartalmazhat perjelet vagy pontot")
 	}
+	// RESERVED NAMESPACE (R-7b). The backup engines key the shares source by the pseudo-stack „_shares"
+	// — a restic tag, a tier-2 dest root and a status record. nbNameRe below starts with [A-Za-z0-9_],
+	// so before this guard „_shares" was an ACCEPTED share name and the underscore namespace was not in
+	// fact reserved (the R-7b task's assumption to the contrary was verified false here). Reserving the
+	// whole leading-underscore space keeps future system keys collision-free too. Validation runs on
+	// ADD only, so an already-registered share is never invalidated retroactively.
+	if strings.HasPrefix(name, "_") {
+		return fmt.Errorf("a megosztás neve nem kezdődhet aláhúzással — ezek a nevek a rendszernek vannak fenntartva")
+	}
 	if !nbNameRe.MatchString(name) {
 		return fmt.Errorf("a megosztás neve csak betűt, számot, kötőjelet és aláhúzást tartalmazhat")
 	}
diff --git a/controller/internal/stacks/samba.go b/controller/internal/stacks/samba.go
index aaaf116..8e1df19 100644
--- a/controller/internal/stacks/samba.go
+++ b/controller/internal/stacks/samba.go
@@ -22,8 +22,9 @@ import (
 const (
 	// SambaStackName is the stack directory name under StacksDir (and the protected-stack key).
 	SambaStackName = "samba"
-	// sambaContainer is the fixed container name (set in the generated compose).
-	sambaContainer = "felhom-samba"
+	// sambaContainer is the fixed container name. Aliased from the RENDERER's constant so the compose
+	// this package writes and the execs it runs can never name different containers.
+	sambaContainer = infra.SambaContainerName
 	// sambaUID is the household uid/gid every SMB write is forced to, so apps (group 1000) and both
 	// backup tiers see consistent ownership.
 	sambaUID = 1000