v0.55.0: raw-device discovery + registry-sourced drive tracking (Impl-2a)

GET /disks/candidates enumerates host disks the Impl-1 unclaimed filter proves free (init/attach split). RegistryKnownTargets sources the watchdog's known-drive set from the intent registry + Felhom .mount units (not Observe/PVE storages) — decouples drive health from PVE storage (closes the registry-only false-detach class); Observe kept for real PVE storages + a deduped /disks union. Idempotent existing-drive migration at start. Tests + red-proof (Observe misses a registry-only drive; registry provider tracks it). go build/vet/test clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-01 17:33:49 +02:00
parent 066e3bf153
commit 91f6a26490
10 changed files with 488 additions and 5 deletions
@@ -1,3 +1,31 @@
+## v0.55.0 — raw-device discovery + registry-sourced drive tracking (Impl-2a) (2026-07-01)
+
+Agent backend for drive enrollment (SPIKE-drive-enrollment §SQ1/SQ4/SQ5). Makes raw (non-PVE-storage)
+drives (a) discoverable for enrollment and (b) health-tracked WITHOUT being a PVE storage — so a drive
+enrolled the new way isn't enrolled-but-untracked (the 3b-fix false-detach class). No mkfs here (Impl-1
+owns it); the controller wizard rewiring is Impl-2b.
+
+- **`GET /disks/candidates`** (`internal/storage/candidates.go` + `internal/localapi/disks.go`):
+  enumerates host whole-disks from `/sys/block`, runs the Impl-1 unclaimed filter, and returns the free
+  ones with probe info (size/model/FS/data-bearing/durable-id preview), split into `initialize` (all
+  unclaimed) and `attach` (the subset carrying a mountable ext4/xfs FS). Fail-safe carries through (a
+  device not provably unclaimed is omitted).
+- **`RegistryKnownTargets`** (`internal/storage/registry_known.go`): the watchdog's known-DRIVE set now
+  comes from the intent registry + Felhom `.mount` units, NOT `Observe()` (PVE storages). A unit is
+  tracked iff its intent ≠ `new` (enrolled/ejected/decommissioned; the watchdog's existing IntentReader
+  gate still decides re-mount). `main.go` swaps the watchdog `KnownTargets` source. **`Observe()` is
+  KEPT** for real PVE storages (local/local-lvm/pbs) — reports + the `/disks` view.
+- **`handleDisks` union:** additive + deduped-by-mount-path — appends registry drives Observe doesn't
+  surface (a registry-only drive now appears in the agent-view) without dropping any Observe row (can't
+  regress the current view).
+- **Existing-drive migration** (`ReconcileExistingDrives`, idempotent, at agent start): records each
+  currently-mounted Felhom-unit drive as `enrolled` so the registry-sourced `Known()` tracks it without
+  its legacy PVE dir-storage. Does NOT create/remove PVE storages.
+- **Tests:** RegistryKnownTargets (enrolled tracked / new excluded / ejected tracked) + the **red-proof**
+  (Observe-based `Known()` misses a drive with no PVE storage; the registry provider tracks it);
+  migration idempotency; candidates init/attach split. `go build`/`vet`/`test ./...` clean. Watchdog/
+  HostLiveness/Remounter unchanged (only the source injected).
+
 ## v0.54.0 — format-safety foundation: unclaimed-disk guard + guarded-mkfs wrapper (2026-07-01)

 Impl-1 (SPIKE-drive-enrollment-2026-07-01). Hardens the destructive `Format`/mkfs path BEFORE the
@@ -250,6 +250,12 @@ of `/` (blast-radius containment on a shared host) — the restore is how a fres
 that scoped token (`VM.Allocate`+`Pool.Allocate` at `/pool/felhom`). Layout + validation:
 `felhom.eu/documentation/audits/SPIKE-pool-scoped-acl-2026-07-01.md`.

+**Drive discovery + tracking (v0.55.0, Impl-2a):** `GET /disks/candidates` lists host disks the Impl-1
+filter proves are free to enroll (split initialize/attach). The watchdog's known-drive set is sourced
+from the intent registry + Felhom `.mount` units (`RegistryKnownTargets`), NOT PVE storages — so a drive
+enrolled with no PVE dir-storage is still health-tracked (`Observe()` stays for real PVE storages). The
+controller wizard consuming this is Impl-2b.
+
 **Format safety (v0.54.0, Impl-1):** `Format` (mkfs) is gated by a mandatory **unclaimed-disk guard**
 (`internal/storage/claim.go`) — it refuses any device not provably free for Felhom (OS disk, LVM PV,
 ZFS/mdraid member, foreign mount, read-only; fail-safe on any read error), independent of `DataBearing`.
@@ -45,7 +45,7 @@ import (

 // version is the agent version. Overridable at build time with
 // -ldflags "-X main.version=<v>"; defaults to the in-repo CHANGELOG version.
-var version = "0.54.0"
+var version = "0.55.0"

 // runGuestHook is the PVE pre-start hook body (`felhom-agent guest-hook <vmid> <phase>`). On the
 // pre-start phase it creates placeholder dirs for any absent bind-mount source so the guest always boots
@@ -429,8 +429,23 @@ func runDaemon(cfg config.Config, logger *slog.Logger) int {
 		logger.Warn("storage: format-job store unavailable — format restart-recovery disabled (F20-BUG3)", "err", fjErr)
 		formatJobStore = nil
 	}
+	// Impl-2a: source the watchdog's known-DRIVE set from the intent registry + Felhom .mount units
+	// (NOT PVE storages via Observe) — a registry-only drive (no PVE dir-storage) is now health-tracked.
+	// Real PVE storages (local/local-lvm/pbs) stay Observe-sourced for /disks + reports (observer below).
+	driveUnitDir := cfg.Privileged.UnitDir
+	if driveUnitDir == "" {
+		driveUnitDir = "/etc/systemd/system"
+	}
+	driveKnown := storage.NewRegistryKnownTargets(driveUnitDir, intentReader, logger)
+	// One-time migration: record pre-Impl-2a drives (mounted, with a Felhom unit) as enrolled so the
+	// registry-sourced Known() tracks them without their legacy PVE dir-storage. Idempotent.
+	if mounts, merr := hostReader.Mounts(); merr == nil {
+		storage.ReconcileExistingDrives(driveUnitDir, mounts, intentStore, logger)
+	} else {
+		logger.Warn("storage: existing-drive migration skipped — mount read failed", "err", merr)
+	}
 	watchdog := storage.NewWatchdog(storage.WatchdogOptions{
-		Targets:   storage.NewCachingKnownTargets(observer, cfg.Storage.KnownRefresh()),
+		Targets:   storage.NewCachingKnownTargets(driveKnown, cfg.Storage.KnownRefresh()),
 		Liveness:  storage.NewHostLiveness(hostReader, 0),
 		Remounter: remounter,
 		Intent:    intentReader, // P3: only re-mount enrolled drives
@@ -513,7 +528,7 @@ func runDaemon(cfg config.Config, logger *slog.Logger) int {
 	jobsRunner := signedjobs.NewRunner(client, gate, signedjobs.ExecutorChain{wipeExec, decommExec}, cfg.Hub.HostID, logger)
 	loop.SetEnvelopeObserver(hub.MultiObserver(desiredSyncer, jobsRunner))

-	localSrv := buildLocalAPIServer(cfg, px, backupStore, observer, hostOps, gate, collector, intentRec, guestBindStore, formatJobStore, logger, &localTokens)
+	localSrv := buildLocalAPIServer(cfg, px, backupStore, observer, driveKnown, hostOps, gate, collector, intentRec, guestBindStore, formatJobStore, logger, &localTokens)
 	if localTokens != nil {
 		defer localTokens.Close()
 	}
@@ -706,7 +721,7 @@ func buildRestoreTestScheduler(cfg config.Config, px *proxmox.Client, engine *re
 // leaf (stable fingerprint). Any failure DISABLES the server (returns nil) WITHOUT crashing the
 // daemon — the host still reports/reconciles; only the controller channel is unavailable until
 // fixed. The opened token store is returned via outTokens so the caller can Close it.
-func buildLocalAPIServer(cfg config.Config, px *proxmox.Client, store *backup.Store, observer *storage.Observer, hostOps *storage.SudoHostOps, gate *reconcile.Gate, collector *hub.Collector, intent localapi.IntentRecorder, guestBinds *localapi.GuestBindStore, formatJobs *localapi.FormatJobStore, logger *slog.Logger, outTokens **localapi.TokenStore) *localapi.Server {
+func buildLocalAPIServer(cfg config.Config, px *proxmox.Client, store *backup.Store, observer *storage.Observer, driveTargets storage.KnownTargets, hostOps *storage.SudoHostOps, gate *reconcile.Gate, collector *hub.Collector, intent localapi.IntentRecorder, guestBinds *localapi.GuestBindStore, formatJobs *localapi.FormatJobStore, logger *slog.Logger, outTokens **localapi.TokenStore) *localapi.Server {
 	if !cfg.LocalAPI.Enabled() {
 		return nil
 	}
@@ -751,6 +766,7 @@ func buildLocalAPIServer(cfg config.Config, px *proxmox.Client, store *backup.St
 		Backups:       runner,
 		Store:         store,
 		Storage:       observer,
+		DriveTargets:  driveTargets, // Impl-2a: registry+units drives for the /disks view (union w/ Observe storages)
 		Tokens:        tokens,
 		BackupCadence: cfg.Backup.BackupCadence(),
 		// Disk management (slice 8C): the privileged host surface + the data-bearing wipe gate.
@@ -27,6 +27,9 @@ type DiskOps interface {
 	Unmount(ctx context.Context, where string) error
 	Format(ctx context.Context, device, fstype string) error
 	InspectDevice(ctx context.Context, device string) (storage.DeviceProbe, error)
+	// ListCandidateDisks enumerates host disks the Impl-1 unclaimed filter proves are free to enroll
+	// (Impl-2a discovery). Fail-safe: a device not provably unclaimed is omitted.
+	ListCandidateDisks(ctx context.Context) ([]storage.CandidateDisk, error)
 }

 // StorageGate authorizes a DESTRUCTIVE storage op (a data-bearing wipe/format) through the
@@ -205,6 +208,33 @@ func (s *Server) handleDisks(w http.ResponseWriter, r *http.Request, vmid int) {
 		}
 		out = append(out, di)
 	}
+	// Impl-2a: union in registry+units drives that Observe() does NOT surface (a drive with no PVE
+	// dir-storage). Additive + deduped by mount path — an existing drive already shown via Observe is
+	// NOT duplicated, and no Observe row is dropped (so this can never regress the current view).
+	if s.driveTargets != nil {
+		seen := make(map[string]bool, len(out))
+		for _, d := range out {
+			if d.MountPath != "" {
+				seen[d.MountPath] = true
+			}
+		}
+		if drives, derr := s.driveTargets.Known(r.Context()); derr == nil {
+			for _, d := range drives {
+				if d.MountPath == "" || seen[d.MountPath] {
+					continue
+				}
+				out = append(out, DiskInfo{
+					Name: d.Name, Type: d.Type, State: "attached",
+					MountPath: d.MountPath, DurableID: d.DurableID,
+					Role:          string(storage.RoleUserData),
+					GuestAttached: boundPaths[d.MountPath],
+				})
+			}
+		} else {
+			s.logger.Warn("disks: registry drive union skipped", "err", derr)
+		}
+	}
+
 	// Guest boot-id (intermediary model): changes on every guest boot, stable across controller restarts.
 	// The controller persists it and deterministically recreates drive-backed apps when it changes.
 	bootID := ""
@@ -214,6 +244,31 @@ func (s *Server) handleDisks(w http.ResponseWriter, r *http.Request, vmid int) {
 	writeOK(w, map[string]any{"vmid": vmid, "disks": out, "guest_boot_id": bootID})
 }

+// handleDiskCandidates (Impl-2a) lists the host disks FREE for Felhom to enroll — the raw-device
+// discovery the enrollment wizard (Impl-2b) consumes. Read-only + benign (the Impl-1 filter never
+// offers a claimed/OS disk). Response splits into `initialize` (all unclaimed) and `attach` (the subset
+// already carrying a mountable FS). GET /disks/candidates.
+func (s *Server) handleDiskCandidates(w http.ResponseWriter, r *http.Request, vmid int) {
+	if s.disks == nil {
+		writeErr(w, http.StatusServiceUnavailable, "disk management not configured on this host")
+		return
+	}
+	cands, err := s.disks.ListCandidateDisks(r.Context())
+	if err != nil {
+		writeErr(w, http.StatusBadGateway, "could not scan host disks")
+		return
+	}
+	initialize := make([]storage.CandidateDisk, 0, len(cands))
+	attach := make([]storage.CandidateDisk, 0)
+	for _, c := range cands {
+		initialize = append(initialize, c) // every unclaimed disk can be initialized (format → enroll)
+		if c.Mountable {
+			attach = append(attach, c) // already has a mountable FS → attach without formatting
+		}
+	}
+	writeOK(w, map[string]any{"vmid": vmid, "initialize": initialize, "attach": attach})
+}
+
 type assignRequest struct {
 	VMID    int    `json:"vmid"`
 	UUID    string `json:"uuid"`
@@ -0,0 +1,40 @@
+package localapi
+
+import (
+	"encoding/json"
+	"net/http"
+	"testing"
+
+	"gitea.dooplex.hu/admin/felhom-agent/internal/storage"
+)
+
+// handleDiskCandidates splits discovery into initialize (all unclaimed) + attach (mountable-FS subset).
+func TestDiskCandidates_Split(t *testing.T) {
+	d := &fakeDiskOps{candidates: []storage.CandidateDisk{
+		{Device: "/dev/sdd", SizeBytes: 64 << 30, DataBearing: false},                                    // blank → initialize only
+		{Device: "/dev/sde", FSType: "ext4", DataBearing: true, Mountable: true, MountSource: "/dev/sde1"}, // FS → init + attach
+		{Device: "/dev/sdf", FSType: "ntfs", DataBearing: true, Mountable: false},                         // ntfs → initialize only
+	}}
+	h := newDiskServer(t, d, &fakeGate{}, nil, nil)
+
+	w := do(t, h, "GET", "/disks/candidates", "A", "")
+	if w.Code != http.StatusOK {
+		t.Fatalf("candidates: got %d want 200 (%s)", w.Code, w.Body.String())
+	}
+	var env struct {
+		Data struct {
+			Initialize []storage.CandidateDisk `json:"initialize"`
+			Attach     []storage.CandidateDisk `json:"attach"`
+		} `json:"data"`
+	}
+	if err := json.Unmarshal(w.Body.Bytes(), &env); err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	resp := env.Data
+	if len(resp.Initialize) != 3 {
+		t.Fatalf("initialize must list all 3 unclaimed disks, got %d", len(resp.Initialize))
+	}
+	if len(resp.Attach) != 1 || resp.Attach[0].Device != "/dev/sde" {
+		t.Fatalf("attach must list only the mountable-FS disk, got %+v", resp.Attach)
+	}
+}
@@ -26,6 +26,12 @@ type fakeDiskOps struct {
 	formatCalls  []string
 	mountCalls   []storage.MountSpec
 	unmountCalls []string
+	candidates   []storage.CandidateDisk // returned by ListCandidateDisks
+	candErr      error
+}
+
+func (f *fakeDiskOps) ListCandidateDisks(_ context.Context) ([]storage.CandidateDisk, error) {
+	return f.candidates, f.candErr
 }

 func (f *fakeDiskOps) InspectDevice(_ context.Context, device string) (storage.DeviceProbe, error) {
@@ -70,6 +70,9 @@ type Options struct {
 	Backups    BackupService
 	Store      BackupStore
 	Storage    StorageView
+	// DriveTargets (Impl-2a, optional) yields registry+units-sourced drives for the /disks view, so a
+	// drive with NO PVE storage still appears. Unioned with Storage.Observe (deduped by mount path).
+	DriveTargets storage.KnownTargets
 	Tokens     TokenAuthority
 	// BackupCadence is the per-guest backup interval driving GET /backup/due (slice 8B). A guest
 	// is "due" when no successful backup is recorded OR the newest one is older than this. 0 → a
@@ -154,7 +157,8 @@ type Server struct {
 	guests  GuestAPI
 	backups BackupService
 	store   BackupStore
-	storage StorageView
+	storage      StorageView
+	driveTargets storage.KnownTargets // Impl-2a: registry+units drives for /disks (optional)
 	tokens  TokenAuthority
 	cadence time.Duration
 	logger  *slog.Logger
@@ -226,6 +230,7 @@ func NewServer(o Options) (*Server, error) {
 		backups:      o.Backups,
 		store:        o.Store,
 		storage:      o.Storage,
+		driveTargets: o.DriveTargets,
 		tokens:       o.Tokens,
 		cadence:      cadence,
 		logger:       o.Logger,
@@ -270,6 +275,7 @@ func (s *Server) Handler() http.Handler {
 	mux.HandleFunc("GET /host/metrics", s.withGuest(s.handleHostMetrics))
 	// Disk management (slice 8C) — self-scoped; format routes through the data-bearing classifier+gate.
 	mux.HandleFunc("GET /disks", s.withGuest(s.handleDisks))
+	mux.HandleFunc("GET /disks/candidates", s.withGuest(s.handleDiskCandidates))
 	mux.HandleFunc("POST /disks/assign", s.withGuest(s.handleDiskAssign))
 	mux.HandleFunc("POST /disks/eject", s.withGuest(s.handleDiskEject))
 	mux.HandleFunc("POST /disks/decommission", s.withGuest(s.handleDiskDecommission))
@@ -0,0 +1,111 @@
+package storage
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"strconv"
+	"strings"
+)
+
+// CandidateDisk is one host block device that the Impl-1 unclaimed filter proved is FREE for Felhom to
+// enroll (Impl-2a discovery). The controller wizard (Impl-2b) renders these; enrollment then formats
+// (guarded, Impl-1) and/or mounts + binds. Blank disks are "initialize"-only; disks already carrying a
+// mountable FS are ALSO "attach" candidates.
+type CandidateDisk struct {
+	Device      string `json:"device"`       // whole disk, e.g. /dev/sdd
+	SizeBytes   int64  `json:"size_bytes"`   // 0 when unreadable
+	Model       string `json:"model,omitempty"`
+	FSType      string `json:"fstype,omitempty"`      // first filesystem found ("" = blank)
+	DataBearing bool   `json:"data_bearing"`          // has any FS / partition (→ wipe-confirm downstream)
+	Mountable   bool   `json:"mountable"`             // carries an ext4/xfs FS the agent can mount as-is
+	MountSource string `json:"mount_source,omitempty"` // the node to mount for attach (e.g. /dev/sdd1)
+	DurableID   string `json:"durable_id,omitempty"`   // "uuid:<fs-uuid>" if it has an FS; "" for a blank disk
+}
+
+// mountableFSTypes are the filesystems the agent can mount as-is (EnsureMount) → an "attach" candidate.
+// Others (ntfs, exfat, …) are data-bearing but not attach-mountable → "initialize" only (with wipe).
+var mountableFSTypes = map[string]bool{"ext4": true, "xfs": true}
+
+// ListCandidateDisks enumerates the host's whole disks (from /sys/block — non-privileged) and returns
+// the subset the Impl-1 unclaimed filter proves is free for Felhom. Fail-safe carries through: a device
+// the filter cannot prove unclaimed (incl. any read error) is simply omitted (never offered).
+func (h *SudoHostOps) ListCandidateDisks(ctx context.Context) ([]CandidateDisk, error) {
+	entries, err := os.ReadDir("/sys/block")
+	if err != nil {
+		return nil, err
+	}
+	var out []CandidateDisk
+	for _, e := range entries {
+		name := e.Name()
+		dev := "/dev/" + name
+		// Only real whole disks (sd*/nvme*n*/vd*/hd*). wholeDiskOf rejects loop/ram/dm-/zram/md/sr etc.
+		if wd, ok := wholeDiskOf(dev); !ok || wd != dev {
+			continue
+		}
+		unclaimed, _ := classifyClaim(h.gatherClaimFacts(ctx, dev))
+		if !unclaimed {
+			continue
+		}
+		out = append(out, h.buildCandidate(dev, name))
+	}
+	return out, nil
+}
+
+// buildCandidate assembles the probe info for an already-unclaimed disk. It re-reads the lsblk tree for
+// per-node FS (via the same allowlisted command the claim gather uses) and /sys for size/model.
+func (h *SudoHostOps) buildCandidate(dev, name string) CandidateDisk {
+	c := CandidateDisk{Device: dev}
+	c.SizeBytes = readSysBlockSize(name)
+	c.Model = readSysBlockModel(name)
+
+	// Per-node FS from lsblk (reuses the FELHOM_FORMAT-allowlisted command).
+	lout, _, lerr := h.runner.Run(context.Background(), h.bins.Lsblk, "-J", "-o", "NAME,FSTYPE,PTTYPE,MOUNTPOINT", dev)
+	if lerr == nil {
+		if nodes, perr := parseLsblkNodes(lout); perr == nil {
+			for _, n := range nodes {
+				if n.fstype == "" {
+					continue
+				}
+				c.DataBearing = true
+				if c.FSType == "" {
+					c.FSType = n.fstype // first filesystem found (partition or whole-disk)
+				}
+				if mountableFSTypes[n.fstype] && c.MountSource == "" {
+					c.Mountable = true
+					c.MountSource = "/dev/" + n.name
+					if uuid, ok := h.host.ResolveUUID("/dev/" + n.name); ok {
+						c.DurableID = "uuid:" + uuid
+					}
+				}
+			}
+			// A partition table with no FS is still data-bearing (must be wiped before init).
+			if len(nodes) > 1 {
+				c.DataBearing = true
+			}
+		}
+	}
+	return c
+}
+
+// readSysBlockSize returns the device size in bytes from /sys/block/<name>/size (512-byte sectors).
+func readSysBlockSize(name string) int64 {
+	b, err := os.ReadFile(filepath.Join("/sys/block", name, "size"))
+	if err != nil {
+		return 0
+	}
+	sectors, err := strconv.ParseInt(strings.TrimSpace(string(b)), 10, 64)
+	if err != nil {
+		return 0
+	}
+	return sectors * 512
+}
+
+// readSysBlockModel returns the device model from /sys/block/<name>/device/model ("" if unreadable).
+func readSysBlockModel(name string) string {
+	b, err := os.ReadFile(filepath.Join("/sys/block", name, "device", "model"))
+	if err != nil {
+		return ""
+	}
+	return strings.TrimSpace(string(b))
+}
@@ -0,0 +1,117 @@
+package storage
+
+import (
+	"context"
+	"log/slog"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"gitea.dooplex.hu/admin/felhom-agent/internal/hub"
+)
+
+// RegistryKnownTargets sources the watchdog's known-drive set from the drive INTENT registry + the
+// Felhom `.mount` units — NOT from PVE storages (Observe). Impl-2a decoupling: a drive tracked this way
+// needs NO PVE dir-storage (which is what hid registry-only drives from health/detect, the 3b-fix class).
+// Real PVE storages (local/local-lvm/pbs) are NOT drives and are handled elsewhere via Observe.
+//
+// A unit is included iff its drive's intent is not `new` (enrolled/ejected/decommissioned are all
+// "known" = health-tracked; the watchdog's IntentReader gate then decides whether to RE-MOUNT — only
+// `enrolled` is remounted). A `new`-intent unit (should not normally exist) is excluded so it never
+// auto-mounts.
+type RegistryKnownTargets struct {
+	unitDir string
+	intent  IntentReader
+	logger  *slog.Logger
+}
+
+// NewRegistryKnownTargets builds the provider. unitDir is the systemd unit dir the Felhom `.mount`
+// units live in (e.g. /etc/systemd/system); intent is the durable-id → intent store (may be nil →
+// every Felhom unit is treated as known, matching the legacy ungated behaviour).
+func NewRegistryKnownTargets(unitDir string, intent IntentReader, logger *slog.Logger) *RegistryKnownTargets {
+	if logger == nil {
+		logger = slog.Default()
+	}
+	return &RegistryKnownTargets{unitDir: unitDir, intent: intent, logger: logger}
+}
+
+// Known enumerates the Felhom drive units under unitDir and returns one KnownTarget per enrolled/
+// tracked drive. It mirrors ReassertEnrolledMounts's enumeration (read dir → parseFelhomMountUnit).
+func (r *RegistryKnownTargets) Known(ctx context.Context) ([]KnownTarget, error) {
+	entries, err := os.ReadDir(r.unitDir)
+	if err != nil {
+		return nil, err // let the caching layer keep the last good set; the watchdog logs + skips a cycle
+	}
+	var out []KnownTarget
+	for _, e := range entries {
+		if !strings.HasSuffix(e.Name(), ".mount") {
+			continue
+		}
+		data, rerr := os.ReadFile(filepath.Join(r.unitDir, e.Name()))
+		if rerr != nil {
+			continue
+		}
+		spec, ok := parseFelhomMountUnit(string(data))
+		if !ok {
+			continue // not one of our by-uuid drive units (netmount units are excluded by the parser)
+		}
+		durableID := "uuid:" + spec.UUID // the same scheme enroll records in the intent store
+		if r.intent != nil && r.intent.Get(durableID) == IntentNew {
+			// A unit with no recorded intent — not enrolled/ejected/decommissioned. Do NOT track it
+			// (and the watchdog must not auto-mount it): a genuine drive must be enrolled first.
+			continue
+		}
+		out = append(out, KnownTarget{
+			Name:        spec.Name,
+			Type:        hub.StorageTypeUSB, // a Felhom drive; MountBacked is what the watchdog keys on
+			DurableID:   durableID,
+			UUID:        spec.UUID,
+			MountBacked: true,
+			MountPath:   spec.Where,
+			// BackingDevice left "" — HostLiveness resolves the device by UUID (by-uuid symlink).
+		})
+	}
+	return out, nil
+}
+
+// ReconcileExistingDrives migrates drives enrolled before Impl-2a into the intent-registry model so the
+// registry-sourced Known() tracks them without depending on their (legacy) PVE dir-storage: for each
+// Felhom `.mount` unit whose drive is currently mounted, ensure the intent registry records it
+// `enrolled`. Idempotent (an already-enrolled drive is a no-op) and non-destructive — it creates no PVE
+// storage and removes nothing. Best-effort per drive. Call once at agent start.
+func ReconcileExistingDrives(unitDir string, mounts []Mount, intent *IntentStore, logger *slog.Logger) {
+	if intent == nil || logger == nil {
+		return
+	}
+	mounted := make(map[string]bool, len(mounts))
+	for _, m := range mounts {
+		mounted[m.MountPoint] = true
+	}
+	entries, err := os.ReadDir(unitDir)
+	if err != nil {
+		logger.Warn("migrate: cannot read unit dir — existing-drive reconcile skipped", "dir", unitDir, "err", err)
+		return
+	}
+	for _, e := range entries {
+		if !strings.HasSuffix(e.Name(), ".mount") {
+			continue
+		}
+		data, rerr := os.ReadFile(filepath.Join(unitDir, e.Name()))
+		if rerr != nil {
+			continue
+		}
+		spec, ok := parseFelhomMountUnit(string(data))
+		if !ok || !mounted[spec.Where] {
+			continue // only migrate currently-active Felhom drives
+		}
+		durableID := "uuid:" + spec.UUID
+		if intent.Get(durableID) != IntentNew {
+			continue // already tracked (enrolled/ejected/decommissioned) — idempotent no-op
+		}
+		if serr := intent.SetEnrolled(durableID); serr != nil {
+			logger.Warn("migrate: could not record enrolled intent", "drive", spec.Name, "err", serr)
+		} else {
+			logger.Info("migrate: recorded existing drive as enrolled", "drive", spec.Name, "durable_id", durableID)
+		}
+	}
+}
@@ -0,0 +1,98 @@
+package storage
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"testing"
+)
+
+// fakeIntent is a minimal IntentReader for the registry tests.
+type fakeIntent struct{ m map[string]DriveIntent }
+
+func (f *fakeIntent) Get(durableID string) DriveIntent { return f.m[durableID] }
+
+func writeUnit(t *testing.T, dir, name, content string) {
+	t.Helper()
+	if err := os.WriteFile(filepath.Join(dir, name), []byte(content), 0o644); err != nil {
+		t.Fatalf("write unit %s: %v", name, err)
+	}
+}
+
+func TestRegistryKnownTargets(t *testing.T) {
+	dir := t.TempDir()
+	writeUnit(t, dir, "mnt-a.mount", renderMountUnit(MountSpec{Name: "felhom-a", UUID: "UUID-A", Where: "/mnt/felhom-a", FSType: "ext4"}))
+	writeUnit(t, dir, "mnt-b.mount", renderMountUnit(MountSpec{Name: "felhom-b", UUID: "UUID-B", Where: "/mnt/felhom-b", FSType: "ext4"}))
+	writeUnit(t, dir, "other.mount", "[Mount]\nWhat=/dev/x\nWhere=/mnt/x\n") // not one of ours → ignored
+
+	// felhom-a enrolled, felhom-b has NO intent (new).
+	intent := &fakeIntent{m: map[string]DriveIntent{"uuid:UUID-A": IntentEnrolled}}
+	r := NewRegistryKnownTargets(dir, intent, quietLogger())
+
+	got, err := r.Known(context.Background())
+	if err != nil {
+		t.Fatalf("Known: %v", err)
+	}
+	if len(got) != 1 || got[0].Name != "felhom-a" {
+		t.Fatalf("want only felhom-a (enrolled); new felhom-b excluded, non-felhom ignored — got %+v", got)
+	}
+	kt := got[0]
+	if kt.DurableID != "uuid:UUID-A" || kt.UUID != "UUID-A" || !kt.MountBacked || kt.MountPath != "/mnt/felhom-a" {
+		t.Fatalf("KnownTarget fields wrong: %+v", kt)
+	}
+
+	// An EJECTED drive is still tracked (known) — the watchdog's IntentReader gate prevents remount.
+	intent.m["uuid:UUID-B"] = IntentEjected
+	got2, _ := r.Known(context.Background())
+	if len(got2) != 2 {
+		t.Fatalf("ejected drive must still be a Known target: %+v", got2)
+	}
+}
+
+// The decoupling red-proof: with NO PVE storage for a drive, the OLD Observe-based Known() misses it,
+// while the registry+units provider tracks it. This is exactly the 3b-fix class the decoupling closes.
+func TestRegistryVsObserve_RedProof(t *testing.T) {
+	dir := t.TempDir()
+	writeUnit(t, dir, "mnt-a.mount", renderMountUnit(MountSpec{Name: "felhom-a", UUID: "UUID-A", Where: "/mnt/felhom-a", FSType: "ext4"}))
+	intent := &fakeIntent{m: map[string]DriveIntent{"uuid:UUID-A": IntentEnrolled}}
+
+	// OLD path: Observer with an API that has NO storages → Known() is empty (drive invisible).
+	obs := NewObserver(&fakeStorageAPI{node: "n"}, &fakeHostReader{}, nil, quietLogger())
+	oldKnown, _ := obs.Known(context.Background())
+	if len(oldKnown) != 0 {
+		t.Fatalf("precondition: Observe-based Known should be empty with no PVE storage, got %+v", oldKnown)
+	}
+
+	// NEW path: the registry provider tracks the drive from unit + intent, no PVE storage needed.
+	newKnown, _ := NewRegistryKnownTargets(dir, intent, quietLogger()).Known(context.Background())
+	if len(newKnown) != 1 || newKnown[0].Name != "felhom-a" {
+		t.Fatalf("registry provider must track the registry-only drive: %+v", newKnown)
+	}
+}
+
+func TestReconcileExistingDrives_Idempotent(t *testing.T) {
+	dir := t.TempDir()
+	writeUnit(t, dir, "mnt-a.mount", renderMountUnit(MountSpec{Name: "felhom-a", UUID: "UUID-A", Where: "/mnt/felhom-a", FSType: "ext4"}))
+	store, err := OpenIntentStore(filepath.Join(t.TempDir(), "intents.json"))
+	if err != nil {
+		t.Fatalf("intent store: %v", err)
+	}
+	mounts := []Mount{{Device: "/dev/disk/by-uuid/UUID-A", MountPoint: "/mnt/felhom-a"}}
+
+	// First reconcile: an un-recorded mounted drive → becomes enrolled.
+	ReconcileExistingDrives(dir, mounts, store, quietLogger())
+	if store.Get("uuid:UUID-A") != IntentEnrolled {
+		t.Fatalf("reconcile must record the mounted drive enrolled, got %q", store.Get("uuid:UUID-A"))
+	}
+	// Idempotent: a re-run doesn't change/downgrade it.
+	ReconcileExistingDrives(dir, mounts, store, quietLogger())
+	if store.Get("uuid:UUID-A") != IntentEnrolled {
+		t.Fatalf("re-run must be a no-op, got %q", store.Get("uuid:UUID-A"))
+	}
+	// An UNMOUNTED drive is NOT auto-enrolled by the migration.
+	writeUnit(t, dir, "mnt-c.mount", renderMountUnit(MountSpec{Name: "felhom-c", UUID: "UUID-C", Where: "/mnt/felhom-c", FSType: "ext4"}))
+	ReconcileExistingDrives(dir, mounts, store, quietLogger())
+	if store.Get("uuid:UUID-C") != IntentNew {
+		t.Fatalf("unmounted drive must NOT be migrated, got %q", store.Get("uuid:UUID-C"))
+	}
+}