feat(volsnap): volume snapshot restore (backlog #6)
Restore a captured volume snapshot onto an image workload's live host-bind
data volumes, then redeploy — the most destructive workload action, built to
the adversarially-reviewed design (C1–C6) with all data-loss guards.
- Engine.Restore (engine-owned): all-or-nothing pre-flight re-resolution from
the workload's CURRENT config (never the tamperable manifest), per-filesystem
disk pre-check, per-workload lock, container quiesce, extract-to-tmp, durable
pre-restore snapshot, write-ahead journal, atomic rename swap, redeploy, and
crash-recovery sweep (RecoverInterruptedRestores) wired before serving.
- internal/keyedmutex: shared per-key lock; deployer now serializes every
deploy entrypoint per workload via DispatchPlugin (+ LockWorkload/RedeployLocked
for the restore re-dispatch, no deadlock).
- Untrusted-archive extractor: zip-slip containment, type allow-list (reg/dir
only), decompression-bomb cap, manifest-index bounds.
- POST /api/workloads/{id}/snapshots/{sid}/restore: admin, X-Confirm-Restore
header (CSRF), per-workload single-flight (409).
- WebUI: Restore button + danger ConfirmDialog + busy state + i18n (en/ru).
Scope: image-source only; scopes absolute/stage/project (driven off the same
supportedScopes constant capture uses).
Plan-reviewed before coding; per-phase go/security/ts reviews; final review
READY TO MERGE. Security review caught + fixed a CRITICAL manifest-Source path
traversal (re-derive target from current config + base containment).
Plan: plans/volume-snapshot-restore/
This commit is contained in:
@@ -14,6 +14,7 @@ import (
|
||||
"github.com/alexei/tinyforge/internal/dns"
|
||||
"github.com/alexei/tinyforge/internal/docker"
|
||||
"github.com/alexei/tinyforge/internal/events"
|
||||
"github.com/alexei/tinyforge/internal/keyedmutex"
|
||||
"github.com/alexei/tinyforge/internal/notify"
|
||||
"github.com/alexei/tinyforge/internal/npm"
|
||||
"github.com/alexei/tinyforge/internal/proxy"
|
||||
@@ -56,6 +57,11 @@ type Server struct {
|
||||
// two concurrent syncs can't race on source_config (review S5).
|
||||
gitopsSync keyedMutex
|
||||
|
||||
// volRestoreInFlight is a per-workload single-flight guard for volume
|
||||
// snapshot restore: a concurrent restore of the same workload is rejected
|
||||
// fast with 409 (TryLock) rather than queuing behind the deployer lock.
|
||||
volRestoreInFlight keyedmutex.Mutex
|
||||
|
||||
dnsProviderMu sync.RWMutex
|
||||
dnsProvider dns.Provider
|
||||
onDNSProviderChanged DNSProviderChangedFunc
|
||||
@@ -359,6 +365,10 @@ func (s *Server) Router() chi.Router {
|
||||
r.With(auth.AdminOnly).Get("/snapshots", s.listWorkloadSnapshots)
|
||||
r.With(auth.AdminOnly).Get("/snapshotable", s.getWorkloadSnapshotable)
|
||||
r.With(auth.AdminOnly).Post("/snapshots", s.createWorkloadSnapshot)
|
||||
// Restore overwrites live volume data and restarts the app — the
|
||||
// most destructive workload action. Admin-gated + X-Confirm-Restore
|
||||
// header (CSRF) + per-workload single-flight, mirroring DB restore.
|
||||
r.With(auth.AdminOnly).Post("/snapshots/{sid}/restore", s.restoreWorkloadSnapshot)
|
||||
|
||||
// Runtime view: per-source persisted state + storage usage.
|
||||
// Read-only; safe for any authenticated user.
|
||||
|
||||
@@ -140,6 +140,72 @@ func (s *Server) deleteSnapshot(w http.ResponseWriter, r *http.Request) {
|
||||
respondJSON(w, http.StatusOK, map[string]string{"status": "deleted"})
|
||||
}
|
||||
|
||||
// restoreWorkloadSnapshot handles POST /api/workloads/{id}/snapshots/{sid}/restore.
|
||||
//
|
||||
// This is the most destructive workload action: it overwrites the app's live
|
||||
// volume data with the snapshot and recreates its containers. It is guarded like
|
||||
// the DB restore — admin-only, an X-Confirm-Restore header that must echo the
|
||||
// snapshot id (defeats CSRF form/img posts, which can't set custom headers), and
|
||||
// a per-workload single-flight so a double-click can't stack two restores. All
|
||||
// the dangerous lock/stop/swap/redeploy logic lives in Engine.Restore; this
|
||||
// handler only validates and delegates.
|
||||
func (s *Server) restoreWorkloadSnapshot(w http.ResponseWriter, r *http.Request) {
|
||||
if s.snapshotEngine == nil {
|
||||
respondError(w, http.StatusServiceUnavailable, "snapshot engine not initialized")
|
||||
return
|
||||
}
|
||||
id := chi.URLParam(r, "id")
|
||||
sid := chi.URLParam(r, "sid")
|
||||
|
||||
if confirm := r.Header.Get("X-Confirm-Restore"); confirm != sid {
|
||||
respondError(w, http.StatusBadRequest,
|
||||
"missing or mismatched X-Confirm-Restore header (must equal snapshot id)")
|
||||
return
|
||||
}
|
||||
|
||||
// Up-front validation for precise client errors (Engine.Restore re-checks
|
||||
// ownership + source kind under the lock).
|
||||
snap, err := s.snapshotEngine.Get(sid)
|
||||
if err != nil {
|
||||
respondError(w, http.StatusNotFound, "snapshot not found")
|
||||
return
|
||||
}
|
||||
if snap.WorkloadID != id {
|
||||
respondError(w, http.StatusBadRequest, "snapshot does not belong to this workload")
|
||||
return
|
||||
}
|
||||
row, ok := s.loadWorkload(w, id)
|
||||
if !ok {
|
||||
return
|
||||
}
|
||||
if row.SourceKind != "image" {
|
||||
respondError(w, http.StatusBadRequest, "restore is only supported for image-source workloads")
|
||||
return
|
||||
}
|
||||
|
||||
// Per-workload single-flight: reject a concurrent restore of the SAME
|
||||
// workload with 409 rather than queuing it behind the deployer lock.
|
||||
release, ok := s.volRestoreInFlight.TryLock(id)
|
||||
if !ok {
|
||||
respondError(w, http.StatusConflict, "a restore is already in progress for this workload")
|
||||
return
|
||||
}
|
||||
defer release()
|
||||
|
||||
if err := s.snapshotEngine.Restore(r.Context(), sid, id); err != nil {
|
||||
// Raw error (which can carry resolved host paths) stays in the log; the
|
||||
// client gets a generic message.
|
||||
slog.Error("snapshots: restore failed", "workload", id, "snapshot", sid, "error", err)
|
||||
respondError(w, http.StatusInternalServerError, "restore failed; see server logs")
|
||||
return
|
||||
}
|
||||
respondJSON(w, http.StatusOK, map[string]any{
|
||||
"status": "restored",
|
||||
"workload_id": id,
|
||||
"snapshot_id": sid,
|
||||
})
|
||||
}
|
||||
|
||||
// downloadSnapshot handles GET /api/snapshots/{sid}/download, streaming the
|
||||
// tar.gz archive. The resolved path is containment-checked against the
|
||||
// snapshot directory.
|
||||
|
||||
@@ -1,12 +1,15 @@
|
||||
package api
|
||||
|
||||
import (
|
||||
"context"
|
||||
"io"
|
||||
"net/http"
|
||||
"net/http/httptest"
|
||||
"os"
|
||||
"path/filepath"
|
||||
"sync"
|
||||
"testing"
|
||||
"time"
|
||||
|
||||
"github.com/alexei/tinyforge/internal/auth"
|
||||
"github.com/alexei/tinyforge/internal/store"
|
||||
@@ -53,7 +56,211 @@ func newSnapshotEnv(t *testing.T) (*apiTestEnv, string) {
|
||||
t.Fatalf("update settings: %v", err)
|
||||
}
|
||||
|
||||
return &apiTestEnv{srv: httpsrv, store: st, dispatcher: dispatcher, adminToken: tok.Token, encKey: encKey}, baseVol
|
||||
return &apiTestEnv{srv: httpsrv, store: st, dispatcher: dispatcher, adminToken: tok.Token, encKey: encKey, snapEngine: snapEng}, baseVol
|
||||
}
|
||||
|
||||
// doRestore issues an authenticated restore POST, optionally setting the
|
||||
// X-Confirm-Restore header (pass confirm="" to omit it).
|
||||
func (e *apiTestEnv) doRestore(t *testing.T, workloadID, sid, confirm string) *http.Response {
|
||||
t.Helper()
|
||||
req, err := http.NewRequest(http.MethodPost,
|
||||
e.srv.URL+"/api/workloads/"+workloadID+"/snapshots/"+sid+"/restore", nil)
|
||||
if err != nil {
|
||||
t.Fatalf("new request: %v", err)
|
||||
}
|
||||
req.Header.Set("Authorization", "Bearer "+e.adminToken)
|
||||
if confirm != "" {
|
||||
req.Header.Set("X-Confirm-Restore", confirm)
|
||||
}
|
||||
resp, err := http.DefaultClient.Do(req)
|
||||
if err != nil {
|
||||
t.Fatalf("do request: %v", err)
|
||||
}
|
||||
return resp
|
||||
}
|
||||
|
||||
// okLifecycle is a no-op volsnap.Lifecycle for HTTP-layer happy-path tests; the
|
||||
// deep restore behavior is covered by the volsnap engine tests.
|
||||
type okLifecycle struct{ tag string }
|
||||
|
||||
func (l *okLifecycle) Lock(string) func() { return func() {} }
|
||||
func (l *okLifecycle) StopContainers(context.Context, string) (string, error) { return l.tag, nil }
|
||||
func (l *okLifecycle) Redeploy(context.Context, store.Workload, string) error { return nil }
|
||||
|
||||
func TestRestoreSnapshot_RequiresConfirmHeader(t *testing.T) {
|
||||
e, _ := newSnapshotEnv(t)
|
||||
w, _ := e.store.CreateWorkload(store.Workload{Name: "a", Kind: "project", SourceKind: "image", SourceConfig: `{"image":"x","port":80}`})
|
||||
snap, _ := e.store.CreateVolumeSnapshot(store.VolumeSnapshot{WorkloadID: w.ID, Filename: "f.tar.gz", Manifest: "[]"})
|
||||
|
||||
// Missing header → 400.
|
||||
resp := e.doRestore(t, w.ID, snap.ID, "")
|
||||
if resp.StatusCode != http.StatusBadRequest {
|
||||
t.Fatalf("missing header status = %d, want 400", resp.StatusCode)
|
||||
}
|
||||
resp.Body.Close()
|
||||
// Mismatched header → 400.
|
||||
resp = e.doRestore(t, w.ID, snap.ID, "not-the-sid")
|
||||
if resp.StatusCode != http.StatusBadRequest {
|
||||
t.Fatalf("mismatched header status = %d, want 400", resp.StatusCode)
|
||||
}
|
||||
resp.Body.Close()
|
||||
}
|
||||
|
||||
func TestRestoreSnapshot_WrongWorkload(t *testing.T) {
|
||||
e, _ := newSnapshotEnv(t)
|
||||
w, _ := e.store.CreateWorkload(store.Workload{Name: "a", Kind: "project", SourceKind: "image", SourceConfig: `{"image":"x","port":80}`})
|
||||
snap, _ := e.store.CreateVolumeSnapshot(store.VolumeSnapshot{WorkloadID: w.ID, Filename: "f.tar.gz", Manifest: "[]"})
|
||||
|
||||
resp := e.doRestore(t, "some-other-workload", snap.ID, snap.ID)
|
||||
if resp.StatusCode != http.StatusBadRequest {
|
||||
t.Fatalf("cross-workload restore status = %d, want 400", resp.StatusCode)
|
||||
}
|
||||
resp.Body.Close()
|
||||
}
|
||||
|
||||
func TestRestoreSnapshot_NonImageWorkload(t *testing.T) {
|
||||
e, _ := newSnapshotEnv(t)
|
||||
w, _ := e.store.CreateWorkload(store.Workload{Name: "site", Kind: "project", SourceKind: "static", SourceConfig: `{}`})
|
||||
snap, _ := e.store.CreateVolumeSnapshot(store.VolumeSnapshot{WorkloadID: w.ID, Filename: "f.tar.gz", Manifest: "[]"})
|
||||
|
||||
resp := e.doRestore(t, w.ID, snap.ID, snap.ID)
|
||||
if resp.StatusCode != http.StatusBadRequest {
|
||||
t.Fatalf("non-image restore status = %d, want 400", resp.StatusCode)
|
||||
}
|
||||
resp.Body.Close()
|
||||
}
|
||||
|
||||
func TestRestoreSnapshot_NotFound(t *testing.T) {
|
||||
e, _ := newSnapshotEnv(t)
|
||||
w, _ := e.store.CreateWorkload(store.Workload{Name: "a", Kind: "project", SourceKind: "image", SourceConfig: `{"image":"x","port":80}`})
|
||||
|
||||
resp := e.doRestore(t, w.ID, "missing-sid", "missing-sid")
|
||||
if resp.StatusCode != http.StatusNotFound {
|
||||
t.Fatalf("unknown snapshot status = %d, want 404", resp.StatusCode)
|
||||
}
|
||||
resp.Body.Close()
|
||||
}
|
||||
|
||||
func TestRestoreSnapshot_HappyPath(t *testing.T) {
|
||||
e, baseVol := newSnapshotEnv(t)
|
||||
e.snapEngine.SetLifecycle(&okLifecycle{tag: "v1"})
|
||||
|
||||
w, err := e.store.CreateWorkload(store.Workload{
|
||||
Name: "data-app", Kind: "project", SourceKind: "image",
|
||||
SourceConfig: `{"image":"reg/app","port":80,"volumes":[{"source":"data","target":"/data","scope":"project"}]}`,
|
||||
})
|
||||
if err != nil {
|
||||
t.Fatalf("create workload: %v", err)
|
||||
}
|
||||
if _, err := e.store.SetWorkloadVolume(store.WorkloadVolume{WorkloadID: w.ID, Target: "/data", Source: "data", Scope: "project"}); err != nil {
|
||||
t.Fatalf("set volume: %v", err)
|
||||
}
|
||||
id8 := w.ID
|
||||
if len(id8) > 8 {
|
||||
id8 = id8[:8]
|
||||
}
|
||||
hostDir := filepath.Join(baseVol, "data-app-"+id8, "data")
|
||||
if err := os.MkdirAll(hostDir, 0o755); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if err := os.WriteFile(filepath.Join(hostDir, "payload.txt"), []byte("ORIGINAL"), 0o644); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
settings, _ := e.store.GetSettings()
|
||||
snap, err := e.snapEngine.Create(w, settings, "base")
|
||||
if err != nil {
|
||||
t.Fatalf("create snapshot: %v", err)
|
||||
}
|
||||
// Drift the live data, then restore.
|
||||
if err := os.WriteFile(filepath.Join(hostDir, "payload.txt"), []byte("CHANGED"), 0o644); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
|
||||
resp := e.doRestore(t, w.ID, snap.ID, snap.ID)
|
||||
if resp.StatusCode != http.StatusOK {
|
||||
body, _ := io.ReadAll(resp.Body)
|
||||
resp.Body.Close()
|
||||
t.Fatalf("restore status = %d, body=%s", resp.StatusCode, body)
|
||||
}
|
||||
resp.Body.Close()
|
||||
if got, _ := os.ReadFile(filepath.Join(hostDir, "payload.txt")); string(got) != "ORIGINAL" {
|
||||
t.Errorf("payload.txt = %q, want ORIGINAL (restored)", got)
|
||||
}
|
||||
}
|
||||
|
||||
// blockingLifecycle blocks in Lock until released, signaling when entered — so
|
||||
// a test can hold one restore in-flight and assert a second is rejected 409.
|
||||
type blockingLifecycle struct {
|
||||
entered chan struct{}
|
||||
release chan struct{}
|
||||
once sync.Once
|
||||
}
|
||||
|
||||
func (l *blockingLifecycle) Lock(string) func() {
|
||||
l.once.Do(func() { close(l.entered) })
|
||||
<-l.release
|
||||
return func() {}
|
||||
}
|
||||
func (l *blockingLifecycle) StopContainers(context.Context, string) (string, error) { return "", nil }
|
||||
func (l *blockingLifecycle) Redeploy(context.Context, store.Workload, string) error { return nil }
|
||||
|
||||
// seedRestorable creates an image workload with a project volume + live data and
|
||||
// a captured snapshot, returning the workload and snapshot ids.
|
||||
func seedRestorable(t *testing.T, e *apiTestEnv, baseVol string) (workloadID, snapshotID string) {
|
||||
t.Helper()
|
||||
w, err := e.store.CreateWorkload(store.Workload{
|
||||
Name: "sf-app", Kind: "project", SourceKind: "image",
|
||||
SourceConfig: `{"image":"reg/app","port":80,"volumes":[{"source":"data","target":"/data","scope":"project"}]}`,
|
||||
})
|
||||
if err != nil {
|
||||
t.Fatalf("create workload: %v", err)
|
||||
}
|
||||
id8 := w.ID
|
||||
if len(id8) > 8 {
|
||||
id8 = id8[:8]
|
||||
}
|
||||
hostDir := filepath.Join(baseVol, "sf-app-"+id8, "data")
|
||||
if err := os.MkdirAll(hostDir, 0o755); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if err := os.WriteFile(filepath.Join(hostDir, "f.txt"), []byte("data"), 0o644); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
settings, _ := e.store.GetSettings()
|
||||
snap, err := e.snapEngine.Create(w, settings, "base")
|
||||
if err != nil {
|
||||
t.Fatalf("create snapshot: %v", err)
|
||||
}
|
||||
return w.ID, snap.ID
|
||||
}
|
||||
|
||||
func TestRestoreSnapshot_SingleFlight409(t *testing.T) {
|
||||
e, baseVol := newSnapshotEnv(t)
|
||||
wid, sid := seedRestorable(t, e, baseVol)
|
||||
bl := &blockingLifecycle{entered: make(chan struct{}), release: make(chan struct{})}
|
||||
e.snapEngine.SetLifecycle(bl)
|
||||
|
||||
// Restore #1: passes validation, takes the single-flight, then blocks inside
|
||||
// the engine's Lock.
|
||||
go func() {
|
||||
resp := e.doRestore(t, wid, sid, sid)
|
||||
resp.Body.Close()
|
||||
}()
|
||||
|
||||
select {
|
||||
case <-bl.entered:
|
||||
case <-time.After(3 * time.Second):
|
||||
t.Fatal("first restore never reached the lifecycle lock")
|
||||
}
|
||||
|
||||
// Restore #2 for the same workload must be rejected fast with 409.
|
||||
resp := e.doRestore(t, wid, sid, sid)
|
||||
got := resp.StatusCode
|
||||
resp.Body.Close()
|
||||
close(bl.release) // let #1 finish
|
||||
if got != http.StatusConflict {
|
||||
t.Fatalf("concurrent restore status = %d, want 409", got)
|
||||
}
|
||||
}
|
||||
|
||||
func TestVolumeSnapshots_EndToEnd(t *testing.T) {
|
||||
|
||||
@@ -15,6 +15,7 @@ import (
|
||||
"github.com/alexei/tinyforge/internal/auth"
|
||||
"github.com/alexei/tinyforge/internal/crypto"
|
||||
"github.com/alexei/tinyforge/internal/store"
|
||||
"github.com/alexei/tinyforge/internal/volsnap"
|
||||
"github.com/alexei/tinyforge/internal/webhook"
|
||||
"github.com/alexei/tinyforge/internal/workload/plugin"
|
||||
|
||||
@@ -75,6 +76,7 @@ type apiTestEnv struct {
|
||||
dispatcher *fakeAPIDispatcher
|
||||
adminToken string
|
||||
encKey [32]byte
|
||||
snapEngine *volsnap.Engine // set by newSnapshotEnv; nil otherwise
|
||||
}
|
||||
|
||||
func (e *apiTestEnv) close() { e.srv.Close() }
|
||||
@@ -670,9 +672,9 @@ func TestGetWorkloadChain_ParentSelfChildren(t *testing.T) {
|
||||
|
||||
resp := e.do(t, http.MethodGet, "/api/workloads/"+parentID+"/chain", nil)
|
||||
var got struct {
|
||||
Parent *map[string]any `json:"parent"`
|
||||
Self map[string]any `json:"self"`
|
||||
Children []map[string]any `json:"children"`
|
||||
Parent *map[string]any `json:"parent"`
|
||||
Self map[string]any `json:"self"`
|
||||
Children []map[string]any `json:"children"`
|
||||
}
|
||||
if errMsg := decodeEnvelope(t, resp, &got); errMsg != "" {
|
||||
t.Fatalf("envelope error: %q", errMsg)
|
||||
|
||||
@@ -5,6 +5,7 @@
|
||||
package deployer
|
||||
|
||||
import (
|
||||
"context"
|
||||
"fmt"
|
||||
"log/slog"
|
||||
"sync"
|
||||
@@ -14,9 +15,11 @@ import (
|
||||
"github.com/alexei/tinyforge/internal/docker"
|
||||
"github.com/alexei/tinyforge/internal/events"
|
||||
"github.com/alexei/tinyforge/internal/health"
|
||||
"github.com/alexei/tinyforge/internal/keyedmutex"
|
||||
"github.com/alexei/tinyforge/internal/notify"
|
||||
"github.com/alexei/tinyforge/internal/proxy"
|
||||
"github.com/alexei/tinyforge/internal/store"
|
||||
"github.com/alexei/tinyforge/internal/workload/plugin"
|
||||
)
|
||||
|
||||
// Deployer owns the dependency bundle each Source plugin needs at deploy
|
||||
@@ -49,6 +52,29 @@ type Deployer struct {
|
||||
drainMu sync.Mutex
|
||||
activeWg sync.WaitGroup
|
||||
shuttingDown atomic.Bool
|
||||
|
||||
// workloadLocks serializes deploy-class operations per workload id so two
|
||||
// concurrent mutators of the same workload (a manual deploy, a webhook/
|
||||
// trigger dispatch, a rollback, a promote, OR a volume-snapshot restore)
|
||||
// can never interleave their container/volume changes. Every deploy
|
||||
// entrypoint funnels through DispatchPlugin, so locking there gates them
|
||||
// all at one choke point. This is the per-workload lock activeWg is NOT
|
||||
// (activeWg is a global drain barrier for graceful shutdown).
|
||||
workloadLocks keyedmutex.Mutex
|
||||
}
|
||||
|
||||
// LockWorkload acquires the per-workload deploy lock for an external critical
|
||||
// section (volume-snapshot restore) and returns the release func. The restore
|
||||
// flow holds this across stop→swap→redeploy and redeploys via RedeployLocked
|
||||
// (which does NOT re-acquire it).
|
||||
func (d *Deployer) LockWorkload(id string) func() { return d.workloadLocks.Lock(id) }
|
||||
|
||||
// RedeployLocked re-dispatches w WITHOUT acquiring the per-workload lock,
|
||||
// because the caller (restore) already holds it via LockWorkload. Calling the
|
||||
// normal DispatchPlugin here would deadlock — Go mutexes are not reentrant.
|
||||
// Not for general use.
|
||||
func (d *Deployer) RedeployLocked(ctx context.Context, w plugin.Workload, intent plugin.DeploymentIntent) error {
|
||||
return d.dispatchLocked(ctx, w, intent)
|
||||
}
|
||||
|
||||
// EventPublisher is the interface for publishing events to the event bus.
|
||||
|
||||
@@ -15,6 +15,18 @@ import (
|
||||
// operator enables auto_backup_before_deploy, a pre-deploy Tinyforge DB
|
||||
// snapshot is taken here, after the source resolves and before it runs.
|
||||
func (d *Deployer) DispatchPlugin(ctx context.Context, w plugin.Workload, intent plugin.DeploymentIntent) error {
|
||||
// C1: serialize all deploy-class work per workload. Held across the whole
|
||||
// deploy so a concurrent deploy/rollback/promote/trigger — or a volume
|
||||
// restore (which redeploys via RedeployLocked while holding this) — can
|
||||
// never interleave container changes for the same workload.
|
||||
unlock := d.workloadLocks.Lock(w.ID)
|
||||
defer unlock()
|
||||
return d.dispatchLocked(ctx, w, intent)
|
||||
}
|
||||
|
||||
// dispatchLocked is DispatchPlugin's body, assuming the per-workload lock is
|
||||
// already held. RedeployLocked calls it directly during restore.
|
||||
func (d *Deployer) dispatchLocked(ctx context.Context, w plugin.Workload, intent plugin.DeploymentIntent) error {
|
||||
if err := d.beginDispatch(); err != nil {
|
||||
metrics.DeploysTotal.Inc(w.SourceKind, "rejected_draining")
|
||||
return err
|
||||
@@ -52,6 +64,12 @@ func (d *Deployer) DispatchPlugin(ctx context.Context, w plugin.Workload, intent
|
||||
// Used when a workload is deleted. Tracked via activeWg so Drain() honours
|
||||
// in-progress teardowns just like deploys.
|
||||
func (d *Deployer) DispatchTeardown(ctx context.Context, w plugin.Workload) error {
|
||||
// Teardown mutates the same containers/routes a deploy does, so it takes the
|
||||
// per-workload lock too (C1). Callers tear down distinct workload ids
|
||||
// sequentially (e.g. preview children then parent), never nested, so no
|
||||
// self-deadlock.
|
||||
unlock := d.workloadLocks.Lock(w.ID)
|
||||
defer unlock()
|
||||
if err := d.beginDispatch(); err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
@@ -0,0 +1,48 @@
|
||||
// Package keyedmutex provides a lazily-populated per-key mutex, so a critical
|
||||
// section can be serialized per key (e.g. per workload id) without a global
|
||||
// lock. It is the shared form of the pattern that originated inline in the
|
||||
// GitOps sync handler; the deployer (per-workload deploy serialization) and the
|
||||
// volume-snapshot restore single-flight both use it.
|
||||
package keyedmutex
|
||||
|
||||
import "sync"
|
||||
|
||||
// Mutex hands out one *sync.Mutex per key on demand. The zero value is ready to
|
||||
// use. The internal map only grows (one entry per distinct key ever locked),
|
||||
// which is bounded in practice by the number of workloads.
|
||||
type Mutex struct {
|
||||
mu sync.Mutex
|
||||
m map[string]*sync.Mutex
|
||||
}
|
||||
|
||||
func (k *Mutex) get(key string) *sync.Mutex {
|
||||
k.mu.Lock()
|
||||
defer k.mu.Unlock()
|
||||
if k.m == nil {
|
||||
k.m = make(map[string]*sync.Mutex)
|
||||
}
|
||||
mu, ok := k.m[key]
|
||||
if !ok {
|
||||
mu = &sync.Mutex{}
|
||||
k.m[key] = mu
|
||||
}
|
||||
return mu
|
||||
}
|
||||
|
||||
// Lock blocks until the mutex for key is acquired, then returns its unlock func.
|
||||
func (k *Mutex) Lock(key string) func() {
|
||||
mu := k.get(key)
|
||||
mu.Lock()
|
||||
return mu.Unlock
|
||||
}
|
||||
|
||||
// TryLock attempts to acquire the mutex for key without blocking. On success it
|
||||
// returns the unlock func and true; if the key is already locked it returns nil
|
||||
// and false so the caller can reject (e.g. HTTP 409) instead of queuing.
|
||||
func (k *Mutex) TryLock(key string) (func(), bool) {
|
||||
mu := k.get(key)
|
||||
if !mu.TryLock() {
|
||||
return nil, false
|
||||
}
|
||||
return mu.Unlock, true
|
||||
}
|
||||
@@ -0,0 +1,83 @@
|
||||
package keyedmutex
|
||||
|
||||
import (
|
||||
"sync"
|
||||
"testing"
|
||||
"time"
|
||||
)
|
||||
|
||||
func TestLockSerializesSameKey(t *testing.T) {
|
||||
var m Mutex
|
||||
unlock := m.Lock("a")
|
||||
|
||||
acquired := make(chan struct{})
|
||||
go func() {
|
||||
u := m.Lock("a")
|
||||
close(acquired)
|
||||
u()
|
||||
}()
|
||||
|
||||
select {
|
||||
case <-acquired:
|
||||
t.Fatal("second Lock on the same key acquired while the first was held")
|
||||
case <-time.After(50 * time.Millisecond):
|
||||
// expected: blocked
|
||||
}
|
||||
unlock()
|
||||
select {
|
||||
case <-acquired:
|
||||
// expected: now acquired
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("second Lock did not acquire after release")
|
||||
}
|
||||
}
|
||||
|
||||
func TestLockIndependentKeys(t *testing.T) {
|
||||
var m Mutex
|
||||
unlockA := m.Lock("a")
|
||||
defer unlockA()
|
||||
// A different key must not block.
|
||||
done := make(chan struct{})
|
||||
go func() { u := m.Lock("b"); u(); close(done) }()
|
||||
select {
|
||||
case <-done:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("Lock on an independent key blocked")
|
||||
}
|
||||
}
|
||||
|
||||
func TestTryLock(t *testing.T) {
|
||||
var m Mutex
|
||||
unlock, ok := m.TryLock("a")
|
||||
if !ok {
|
||||
t.Fatal("TryLock should succeed on a free key")
|
||||
}
|
||||
if _, ok := m.TryLock("a"); ok {
|
||||
t.Fatal("TryLock should fail while the key is held")
|
||||
}
|
||||
unlock()
|
||||
u2, ok := m.TryLock("a")
|
||||
if !ok {
|
||||
t.Fatal("TryLock should succeed after release")
|
||||
}
|
||||
u2()
|
||||
}
|
||||
|
||||
func TestConcurrentLockNoRace(t *testing.T) {
|
||||
var m Mutex
|
||||
var wg sync.WaitGroup
|
||||
counter := 0
|
||||
for i := 0; i < 50; i++ {
|
||||
wg.Add(1)
|
||||
go func() {
|
||||
defer wg.Done()
|
||||
u := m.Lock("shared")
|
||||
counter++ // protected by the keyed lock
|
||||
u()
|
||||
}()
|
||||
}
|
||||
wg.Wait()
|
||||
if counter != 50 {
|
||||
t.Errorf("counter = %d, want 50 (lost updates ⇒ lock not serializing)", counter)
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,21 @@
|
||||
//go:build !windows
|
||||
|
||||
package volsnap
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
|
||||
"golang.org/x/sys/unix"
|
||||
)
|
||||
|
||||
// freeDiskBytes returns the bytes available to an unprivileged process on the
|
||||
// filesystem backing path (used by the restore disk pre-check, C5). path must
|
||||
// exist; callers pass the live dir's parent.
|
||||
func freeDiskBytes(path string) (uint64, error) {
|
||||
var st unix.Statfs_t
|
||||
if err := unix.Statfs(path, &st); err != nil {
|
||||
return 0, fmt.Errorf("statfs %s: %w", path, err)
|
||||
}
|
||||
// Bavail is blocks available to non-root; Bsize is the fragment size.
|
||||
return st.Bavail * uint64(st.Bsize), nil
|
||||
}
|
||||
@@ -0,0 +1,24 @@
|
||||
//go:build windows
|
||||
|
||||
package volsnap
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
|
||||
"golang.org/x/sys/windows"
|
||||
)
|
||||
|
||||
// freeDiskBytes returns the bytes available to the caller on the volume backing
|
||||
// path (used by the restore disk pre-check, C5). Windows is the dev platform;
|
||||
// production runs on Linux (see disk_unix.go).
|
||||
func freeDiskBytes(path string) (uint64, error) {
|
||||
p, err := windows.UTF16PtrFromString(path)
|
||||
if err != nil {
|
||||
return 0, fmt.Errorf("encode path %s: %w", path, err)
|
||||
}
|
||||
var freeAvail, total, totalFree uint64
|
||||
if err := windows.GetDiskFreeSpaceEx(p, &freeAvail, &total, &totalFree); err != nil {
|
||||
return 0, fmt.Errorf("GetDiskFreeSpaceEx %s: %w", path, err)
|
||||
}
|
||||
return freeAvail, nil
|
||||
}
|
||||
@@ -31,6 +31,12 @@ type Engine struct {
|
||||
mu sync.Mutex
|
||||
store *store.Store
|
||||
snapDir string
|
||||
|
||||
// lifecycle is the deploy-side seam restore needs (per-workload lock, stop,
|
||||
// redeploy). Wired post-construction via SetLifecycle from the composition
|
||||
// root so volsnap stays decoupled from the deployer/docker packages. nil
|
||||
// until wired; Restore refuses to run without it.
|
||||
lifecycle Lifecycle
|
||||
}
|
||||
|
||||
// New creates the snapshot engine, ensuring the snapshot directory exists.
|
||||
|
||||
@@ -0,0 +1,162 @@
|
||||
package volsnap
|
||||
|
||||
import (
|
||||
"archive/tar"
|
||||
"compress/gzip"
|
||||
"fmt"
|
||||
"io"
|
||||
"os"
|
||||
"path"
|
||||
"path/filepath"
|
||||
"strconv"
|
||||
"strings"
|
||||
)
|
||||
|
||||
// safeExtractIndex extracts the files archived under the integer subdirectory
|
||||
// `index` of a snapshot tar.gz into dest, returning the total bytes written.
|
||||
//
|
||||
// On RESTORE the archive is treated as UNTRUSTED (it may have been downloaded,
|
||||
// hand-edited, or swapped on disk), so this is hardened well beyond what the
|
||||
// capture writer emits:
|
||||
//
|
||||
// - zip-slip: every resolved target must stay within dest (HasPrefix check on
|
||||
// the cleaned absolute path) — a "../" or absolute member is rejected.
|
||||
// - type allow-list: ONLY regular files and directories are materialized;
|
||||
// symlinks, hardlinks, char/block devices, fifos, and sockets are rejected
|
||||
// outright (never created, never followed) — they could redirect a write
|
||||
// outside the volume or smuggle in a device node.
|
||||
// - decompression bomb: a running byte counter is capped at bombCap; the first
|
||||
// byte past the cap aborts the extraction.
|
||||
//
|
||||
// dest must be a fresh staging directory (files are created O_EXCL). The caller
|
||||
// performs the atomic rename-swap of dest onto the live path separately.
|
||||
func safeExtractIndex(archivePath string, index int, dest string, bombCap int64) (int64, error) {
|
||||
f, err := os.Open(archivePath)
|
||||
if err != nil {
|
||||
return 0, fmt.Errorf("open archive: %w", err)
|
||||
}
|
||||
defer f.Close()
|
||||
|
||||
gz, err := gzip.NewReader(f)
|
||||
if err != nil {
|
||||
return 0, fmt.Errorf("gzip reader: %w", err)
|
||||
}
|
||||
defer gz.Close()
|
||||
|
||||
cleanDest, err := filepath.Abs(dest)
|
||||
if err != nil {
|
||||
return 0, fmt.Errorf("resolve dest: %w", err)
|
||||
}
|
||||
if err := os.MkdirAll(cleanDest, 0o700); err != nil {
|
||||
return 0, fmt.Errorf("create dest: %w", err)
|
||||
}
|
||||
|
||||
tr := tar.NewReader(gz)
|
||||
var written int64
|
||||
for {
|
||||
hdr, err := tr.Next()
|
||||
if err == io.EOF {
|
||||
break
|
||||
}
|
||||
if err != nil {
|
||||
return written, fmt.Errorf("read tar: %w", err)
|
||||
}
|
||||
|
||||
// Archive paths are always forward-slash. path.Clean collapses any
|
||||
// "./" / "../" so the prefix and containment checks see a normal form.
|
||||
name := path.Clean(hdr.Name)
|
||||
if name == "manifest.json" {
|
||||
continue
|
||||
}
|
||||
rel, ok := stripIndexPrefix(name, index)
|
||||
if !ok {
|
||||
continue // belongs to a different volume's subtree
|
||||
}
|
||||
|
||||
switch hdr.Typeflag {
|
||||
case tar.TypeReg, tar.TypeDir:
|
||||
// allowed
|
||||
default:
|
||||
return written, fmt.Errorf("archive entry %q has disallowed type %q", hdr.Name, string(hdr.Typeflag))
|
||||
}
|
||||
|
||||
target := cleanDest
|
||||
if rel != "" {
|
||||
target = filepath.Join(cleanDest, filepath.FromSlash(rel))
|
||||
}
|
||||
if !withinDir(cleanDest, target) {
|
||||
return written, fmt.Errorf("archive entry %q escapes destination", hdr.Name)
|
||||
}
|
||||
|
||||
if hdr.Typeflag == tar.TypeDir {
|
||||
if err := os.MkdirAll(target, 0o700); err != nil {
|
||||
return written, fmt.Errorf("mkdir %s: %w", target, err)
|
||||
}
|
||||
continue
|
||||
}
|
||||
|
||||
if err := os.MkdirAll(filepath.Dir(target), 0o700); err != nil {
|
||||
return written, fmt.Errorf("mkdir parent of %s: %w", target, err)
|
||||
}
|
||||
remaining := bombCap - written
|
||||
if remaining <= 0 {
|
||||
return written, fmt.Errorf("archive exceeds decompression cap of %d bytes", bombCap)
|
||||
}
|
||||
out, err := os.OpenFile(target, os.O_CREATE|os.O_EXCL|os.O_WRONLY, 0o600)
|
||||
if err != nil {
|
||||
return written, fmt.Errorf("create %s: %w", target, err)
|
||||
}
|
||||
// LimitReader to remaining+1: if the entry is larger than the cap allows,
|
||||
// the extra byte is copied and written>bombCap trips the guard below.
|
||||
n, copyErr := io.Copy(out, io.LimitReader(tr, remaining+1))
|
||||
closeErr := out.Close()
|
||||
written += n
|
||||
if copyErr != nil {
|
||||
return written, fmt.Errorf("write %s: %w", target, copyErr)
|
||||
}
|
||||
if closeErr != nil {
|
||||
return written, fmt.Errorf("close %s: %w", target, closeErr)
|
||||
}
|
||||
if written > bombCap {
|
||||
return written, fmt.Errorf("archive exceeds decompression cap of %d bytes", bombCap)
|
||||
}
|
||||
}
|
||||
return written, nil
|
||||
}
|
||||
|
||||
// stripIndexPrefix returns the path relative to the "<index>/" subtree and
|
||||
// whether name belongs to it. name=="<index>" (the subtree root) yields ("", true).
|
||||
// The "/" boundary keeps index 1 from matching "10/...".
|
||||
func stripIndexPrefix(name string, index int) (string, bool) {
|
||||
p := strconv.Itoa(index)
|
||||
if name == p {
|
||||
return "", true
|
||||
}
|
||||
if strings.HasPrefix(name, p+"/") {
|
||||
return name[len(p)+1:], true
|
||||
}
|
||||
return "", false
|
||||
}
|
||||
|
||||
// leadingIndex parses the first path segment of an archive entry name as the
|
||||
// volume index. Returns false for manifest.json or any non-integer prefix.
|
||||
func leadingIndex(name string) (int, bool) {
|
||||
seg := name
|
||||
if i := strings.IndexByte(name, '/'); i >= 0 {
|
||||
seg = name[:i]
|
||||
}
|
||||
idx, err := strconv.Atoi(seg)
|
||||
if err != nil {
|
||||
return 0, false
|
||||
}
|
||||
return idx, true
|
||||
}
|
||||
|
||||
// withinDir reports whether target is base itself or lives beneath it. Both
|
||||
// args must already be cleaned absolute paths.
|
||||
func withinDir(base, target string) bool {
|
||||
if target == base {
|
||||
return true
|
||||
}
|
||||
return strings.HasPrefix(target, base+string(filepath.Separator))
|
||||
}
|
||||
@@ -0,0 +1,166 @@
|
||||
package volsnap
|
||||
|
||||
import (
|
||||
"archive/tar"
|
||||
"compress/gzip"
|
||||
"os"
|
||||
"path/filepath"
|
||||
"strings"
|
||||
"testing"
|
||||
)
|
||||
|
||||
type tentry struct {
|
||||
name string
|
||||
typeflag byte
|
||||
body string
|
||||
linkname string
|
||||
}
|
||||
|
||||
// buildTarGz writes an arbitrary (possibly hostile) tar.gz so the extractor's
|
||||
// untrusted-input hardening can be exercised — writeArchive only emits well-
|
||||
// formed reg/dir entries, which is the wrong shape for these tests.
|
||||
func buildTarGz(t *testing.T, entries []tentry) string {
|
||||
t.Helper()
|
||||
dest := filepath.Join(t.TempDir(), "snap.tar.gz")
|
||||
f, err := os.Create(dest)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
gz := gzip.NewWriter(f)
|
||||
tw := tar.NewWriter(gz)
|
||||
for _, e := range entries {
|
||||
hdr := &tar.Header{Name: e.name, Typeflag: e.typeflag, Mode: 0o600, Linkname: e.linkname}
|
||||
switch e.typeflag {
|
||||
case tar.TypeReg:
|
||||
hdr.Size = int64(len(e.body))
|
||||
case tar.TypeChar, tar.TypeBlock:
|
||||
hdr.Devmajor, hdr.Devminor = 1, 1
|
||||
}
|
||||
if err := tw.WriteHeader(hdr); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if e.typeflag == tar.TypeReg {
|
||||
if _, err := tw.Write([]byte(e.body)); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
}
|
||||
}
|
||||
if err := tw.Close(); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if err := gz.Close(); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if err := f.Close(); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
return dest
|
||||
}
|
||||
|
||||
func TestSafeExtractIndex_RoundTrip(t *testing.T) {
|
||||
arc := buildTarGz(t, []tentry{
|
||||
{name: "0/", typeflag: tar.TypeDir},
|
||||
{name: "0/a.txt", typeflag: tar.TypeReg, body: "hello"},
|
||||
{name: "0/sub/", typeflag: tar.TypeDir},
|
||||
{name: "0/sub/b.txt", typeflag: tar.TypeReg, body: "world"},
|
||||
{name: "manifest.json", typeflag: tar.TypeReg, body: "[]"},
|
||||
})
|
||||
dest := filepath.Join(t.TempDir(), "out")
|
||||
n, err := safeExtractIndex(arc, 0, dest, maxRestoreUncompressedBytes)
|
||||
if err != nil {
|
||||
t.Fatalf("extract: %v", err)
|
||||
}
|
||||
if n != int64(len("hello")+len("world")) {
|
||||
t.Errorf("written = %d, want %d", n, len("hello")+len("world"))
|
||||
}
|
||||
if got, _ := os.ReadFile(filepath.Join(dest, "a.txt")); string(got) != "hello" {
|
||||
t.Errorf("a.txt = %q", got)
|
||||
}
|
||||
if got, _ := os.ReadFile(filepath.Join(dest, "sub", "b.txt")); string(got) != "world" {
|
||||
t.Errorf("sub/b.txt = %q", got)
|
||||
}
|
||||
if _, err := os.Stat(filepath.Join(dest, "manifest.json")); !os.IsNotExist(err) {
|
||||
t.Error("manifest.json must not be extracted into the volume")
|
||||
}
|
||||
}
|
||||
|
||||
func TestSafeExtractIndex_IsolatesIndex(t *testing.T) {
|
||||
arc := buildTarGz(t, []tentry{
|
||||
{name: "1/keep.txt", typeflag: tar.TypeReg, body: "one"},
|
||||
{name: "10/other.txt", typeflag: tar.TypeReg, body: "ten"},
|
||||
{name: "2/nope.txt", typeflag: tar.TypeReg, body: "two"},
|
||||
})
|
||||
dest := filepath.Join(t.TempDir(), "out")
|
||||
if _, err := safeExtractIndex(arc, 1, dest, maxRestoreUncompressedBytes); err != nil {
|
||||
t.Fatalf("extract: %v", err)
|
||||
}
|
||||
if _, err := os.Stat(filepath.Join(dest, "keep.txt")); err != nil {
|
||||
t.Errorf("index 1 file missing: %v", err)
|
||||
}
|
||||
// "10/" must not bleed into index 1 (prefix boundary), nor "2/".
|
||||
for _, leaked := range []string{"other.txt", "nope.txt"} {
|
||||
if _, err := os.Stat(filepath.Join(dest, leaked)); !os.IsNotExist(err) {
|
||||
t.Errorf("index 1 extraction leaked %q from another index", leaked)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func TestSafeExtractIndex_RejectsDisallowedTypes(t *testing.T) {
|
||||
cases := map[string]tentry{
|
||||
"symlink": {name: "0/link", typeflag: tar.TypeSymlink, linkname: "/etc/passwd"},
|
||||
"hardlink": {name: "0/hard", typeflag: tar.TypeLink, linkname: "0/real"},
|
||||
"chardev": {name: "0/cdev", typeflag: tar.TypeChar},
|
||||
"blockdev": {name: "0/bdev", typeflag: tar.TypeBlock},
|
||||
"fifo": {name: "0/fifo", typeflag: tar.TypeFifo},
|
||||
"sparse": {name: "0/sparse", typeflag: tar.TypeCont}, // GNU sparse / contiguous
|
||||
}
|
||||
for name, ent := range cases {
|
||||
t.Run(name, func(t *testing.T) {
|
||||
arc := buildTarGz(t, []tentry{ent})
|
||||
dest := filepath.Join(t.TempDir(), "out")
|
||||
if _, err := safeExtractIndex(arc, 0, dest, maxRestoreUncompressedBytes); err == nil {
|
||||
t.Fatalf("expected %s entry to be rejected", name)
|
||||
}
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
func TestSafeExtractIndex_RejectsBomb(t *testing.T) {
|
||||
arc := buildTarGz(t, []tentry{
|
||||
{name: "0/big.bin", typeflag: tar.TypeReg, body: strings.Repeat("x", 4096)},
|
||||
})
|
||||
dest := filepath.Join(t.TempDir(), "out")
|
||||
if _, err := safeExtractIndex(arc, 0, dest, 1024); err == nil {
|
||||
t.Fatal("expected extraction to abort past the decompression cap")
|
||||
}
|
||||
}
|
||||
|
||||
func TestSafeExtractIndex_NoEscapeOutsideDest(t *testing.T) {
|
||||
// A "../" climb and an absolute member must never materialize a file
|
||||
// outside dest, regardless of which guard catches it. Note the backslash
|
||||
// case is platform-split: on Windows `..\winslip.txt` is a real climb that
|
||||
// withinDir rejects; on Linux it is a literal one-segment filename that
|
||||
// stays harmlessly inside dest (no guard fires). Both satisfy containment.
|
||||
arc := buildTarGz(t, []tentry{
|
||||
{name: "0/../../escape.txt", typeflag: tar.TypeReg, body: "pwned"},
|
||||
{name: "/abs-escape.txt", typeflag: tar.TypeReg, body: "pwned"},
|
||||
{name: `0/..\winslip.txt`, typeflag: tar.TypeReg, body: "pwned"},
|
||||
{name: "0/ok.txt", typeflag: tar.TypeReg, body: "fine"},
|
||||
})
|
||||
outParent := t.TempDir()
|
||||
dest := filepath.Join(outParent, "out")
|
||||
// May or may not error depending on platform/guard; the invariant is that
|
||||
// nothing escapes dest.
|
||||
_, _ = safeExtractIndex(arc, 0, dest, maxRestoreUncompressedBytes)
|
||||
|
||||
for _, escaped := range []string{
|
||||
filepath.Join(outParent, "escape.txt"),
|
||||
filepath.Join(filepath.Dir(outParent), "escape.txt"),
|
||||
filepath.Join(outParent, "abs-escape.txt"),
|
||||
filepath.Join(outParent, "winslip.txt"),
|
||||
} {
|
||||
if _, err := os.Stat(escaped); err == nil {
|
||||
t.Errorf("zip-slip escaped to %s", escaped)
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,235 @@
|
||||
package volsnap
|
||||
|
||||
import (
|
||||
"archive/tar"
|
||||
"compress/gzip"
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"io"
|
||||
"os"
|
||||
"path"
|
||||
"path/filepath"
|
||||
|
||||
"github.com/alexei/tinyforge/internal/store"
|
||||
"github.com/alexei/tinyforge/internal/volume"
|
||||
)
|
||||
|
||||
// maxRestoreUncompressedBytes caps the total decompressed size accepted from a
|
||||
// snapshot archive during restore (decompression-bomb defence). 50 GiB is far
|
||||
// above any realistic app data volume while still bounding a hostile archive.
|
||||
const maxRestoreUncompressedBytes int64 = 50 << 30
|
||||
|
||||
// diskFreeHeadroomBytes is extra free space required beyond the extracted size
|
||||
// so a restore never fills the target filesystem to the brim. The live copy is
|
||||
// renamed aside (no new space), so the new allocation is ~the extracted size;
|
||||
// this headroom covers filesystem overhead and metadata.
|
||||
const diskFreeHeadroomBytes int64 = 256 << 20
|
||||
|
||||
// resolvedVol is a manifest volume whose live host path has been re-resolved
|
||||
// against the workload's CURRENT config (all-or-nothing pre-flight, C3).
|
||||
type resolvedVol struct {
|
||||
Index int
|
||||
Target string
|
||||
Scope string
|
||||
LivePath string
|
||||
}
|
||||
|
||||
// parseManifest decodes the snapshot row's manifest JSON ([]SnapshotVolume).
|
||||
func parseManifest(snap store.VolumeSnapshot) ([]SnapshotVolume, error) {
|
||||
var m []SnapshotVolume
|
||||
if err := json.Unmarshal([]byte(snap.Manifest), &m); err != nil {
|
||||
return nil, fmt.Errorf("parse snapshot manifest: %w", err)
|
||||
}
|
||||
if len(m) == 0 {
|
||||
return nil, fmt.Errorf("snapshot manifest is empty")
|
||||
}
|
||||
return m, nil
|
||||
}
|
||||
|
||||
// preflightResolve re-derives every manifest volume's live host path from the
|
||||
// workload's CURRENT config, ALL-OR-NOTHING (C3): if any snapshotted target is
|
||||
// no longer declared, its scope is unsupported, or it can't resolve, it returns
|
||||
// an error and the caller MUST abort BEFORE stopping containers or touching
|
||||
// disk — config drift mid-restore is silent corruption.
|
||||
//
|
||||
// SECURITY: the swap target is keyed on the manifest's container Target path but
|
||||
// its host directory is derived from the CURRENT (trusted, operator-set)
|
||||
// Source/Scope — never from the snapshot manifest's persisted Source/Scope. The
|
||||
// manifest column is attacker-influenceable (e.g. a restored/tampered DB), and
|
||||
// trusting its Source for stage/project scope would let `Source:"../../etc"`
|
||||
// redirect the destructive rename-swap outside the volume tree. As defence in
|
||||
// depth, base-relative resolved paths are asserted to stay under BaseVolumePath.
|
||||
func preflightResolve(st *store.Store, w store.Workload, settings store.Settings, manifest []SnapshotVolume) ([]resolvedVol, error) {
|
||||
current, err := volumesByTarget(st, w)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("load current volumes: %w", err)
|
||||
}
|
||||
params := volume.ResolveWorkloadParams{
|
||||
BasePath: settings.BaseVolumePath,
|
||||
WorkloadID: w.ID,
|
||||
WorkloadName: w.Name,
|
||||
AllowedVolumePaths: settings.AllowedVolumePaths,
|
||||
}
|
||||
out := make([]resolvedVol, 0, len(manifest))
|
||||
for _, mv := range manifest {
|
||||
// A negative index can never name an archive subtree.
|
||||
if mv.Index < 0 {
|
||||
return nil, fmt.Errorf("volume %q has invalid index %d", mv.Target, mv.Index)
|
||||
}
|
||||
cur, ok := current[mv.Target]
|
||||
if !ok {
|
||||
return nil, fmt.Errorf("volume %q is no longer declared by the workload", mv.Target)
|
||||
}
|
||||
if !supportedScopes[cur.Scope] {
|
||||
return nil, fmt.Errorf("volume %q scope %q is not restorable", mv.Target, cur.Scope)
|
||||
}
|
||||
live, err := volume.ResolveWorkloadPath(cur, params)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("resolve volume %q (%s): %w", mv.Target, cur.Scope, err)
|
||||
}
|
||||
// Containment: the destructive swap target must stay inside the volume
|
||||
// root. Base-relative scopes must resolve under BaseVolumePath; absolute
|
||||
// scope is already constrained to AllowedVolumePaths by the resolver.
|
||||
if cur.Scope != string(store.VolumeScopeAbsolute) {
|
||||
contained, cerr := pathWithinBase(settings.BaseVolumePath, live)
|
||||
if cerr != nil || !contained {
|
||||
return nil, fmt.Errorf("resolved path for volume %q escapes the volume root", mv.Target)
|
||||
}
|
||||
}
|
||||
out = append(out, resolvedVol{Index: mv.Index, Target: mv.Target, Scope: cur.Scope, LivePath: live})
|
||||
}
|
||||
return out, nil
|
||||
}
|
||||
|
||||
// pathWithinBase reports whether target resolves to base or a path beneath it.
|
||||
// An empty base is treated as non-containing (refuse rather than allow).
|
||||
func pathWithinBase(base, target string) (bool, error) {
|
||||
if base == "" {
|
||||
return false, nil
|
||||
}
|
||||
absBase, err := filepath.Abs(base)
|
||||
if err != nil {
|
||||
return false, err
|
||||
}
|
||||
absTarget, err := filepath.Abs(target)
|
||||
if err != nil {
|
||||
return false, err
|
||||
}
|
||||
return withinDir(absBase, absTarget), nil
|
||||
}
|
||||
|
||||
// archiveUncompressedSize scans the archive's tar headers and returns the
|
||||
// per-index and total uncompressed sizes, enforcing bombCap so a hostile
|
||||
// archive can't make the disk pre-check allocate unbounded. Feeds the
|
||||
// per-filesystem free-space pre-check (C5).
|
||||
//
|
||||
// The total is a LOWER-BOUND estimate of on-disk consumption: it sums regular-
|
||||
// file bytes only, ignoring directory entries and per-file inode/block-rounding
|
||||
// overhead, so a volume of many tiny files consumes more than reported. The
|
||||
// real safety net is the staged extract + atomic swap (a mid-extract ENOSPC
|
||||
// discards the staging dir and leaves live untouched), not this pre-check.
|
||||
//
|
||||
// "No body copy" is at the API level only — tar.Next still inflates and
|
||||
// discards each skipped body, so a 50 GiB-of-headers archive does 50 GiB of
|
||||
// gzip work; bombCap bounds that.
|
||||
func archiveUncompressedSize(archivePath string, bombCap int64) (perIndex map[int]int64, total int64, err error) {
|
||||
f, err := os.Open(archivePath)
|
||||
if err != nil {
|
||||
return nil, 0, fmt.Errorf("open archive: %w", err)
|
||||
}
|
||||
defer f.Close()
|
||||
gz, err := gzip.NewReader(f)
|
||||
if err != nil {
|
||||
return nil, 0, fmt.Errorf("gzip reader: %w", err)
|
||||
}
|
||||
defer gz.Close()
|
||||
|
||||
perIndex = map[int]int64{}
|
||||
tr := tar.NewReader(gz)
|
||||
for {
|
||||
hdr, e := tr.Next()
|
||||
if e == io.EOF {
|
||||
break
|
||||
}
|
||||
if e != nil {
|
||||
return nil, 0, fmt.Errorf("read tar: %w", e)
|
||||
}
|
||||
if hdr.Typeflag != tar.TypeReg {
|
||||
continue
|
||||
}
|
||||
name := path.Clean(hdr.Name)
|
||||
if name == "manifest.json" {
|
||||
continue
|
||||
}
|
||||
idx, ok := leadingIndex(name)
|
||||
if !ok {
|
||||
continue
|
||||
}
|
||||
total += hdr.Size
|
||||
if total > bombCap {
|
||||
return nil, 0, fmt.Errorf("archive exceeds decompression cap of %d bytes", bombCap)
|
||||
}
|
||||
perIndex[idx] += hdr.Size
|
||||
}
|
||||
return perIndex, total, nil
|
||||
}
|
||||
|
||||
// swap records one volume's atomic dir replacement so it can be rolled back.
|
||||
type swap struct {
|
||||
live string
|
||||
old string // where the prior live dir was set aside ("" if live didn't exist)
|
||||
tmp string // staging dir holding the freshly-extracted data
|
||||
hadOld bool // whether a prior live dir existed and was moved to old
|
||||
}
|
||||
|
||||
// stagingDirs returns the per-volume tmp and old staging paths as SIBLINGS of
|
||||
// the live dir's parent, so every rename in the swap is intra-filesystem and
|
||||
// therefore atomic (R2). A cross-device rename (live is itself a mountpoint)
|
||||
// fails loudly in swapVolumeDir rather than silently degrading to a copy.
|
||||
func stagingDirs(live, token string, index int) (tmp, old string) {
|
||||
parent := filepath.Dir(live)
|
||||
base := fmt.Sprintf(".tf-restore-%s-%d", token, index)
|
||||
return filepath.Join(parent, base+".tmp"), filepath.Join(parent, base+".old")
|
||||
}
|
||||
|
||||
// swapVolumeDir performs the crash-minimal two-rename swap: set the live dir
|
||||
// aside to old (if it exists), then move the staged tmp into place (C2). On the
|
||||
// second rename failing it reverts the first so live is never left missing.
|
||||
// Returns whether a prior live dir was preserved at old (for rollback).
|
||||
func swapVolumeDir(live, tmp, old string) (hadOld bool, err error) {
|
||||
if _, statErr := os.Lstat(live); statErr == nil {
|
||||
if rerr := os.Rename(live, old); rerr != nil {
|
||||
return false, fmt.Errorf("set aside live %s: %w", live, rerr)
|
||||
}
|
||||
hadOld = true
|
||||
} else if !os.IsNotExist(statErr) {
|
||||
return false, fmt.Errorf("stat live %s: %w", live, statErr)
|
||||
}
|
||||
|
||||
if mkErr := os.MkdirAll(filepath.Dir(live), 0o700); mkErr != nil {
|
||||
if hadOld {
|
||||
_ = os.Rename(old, live)
|
||||
}
|
||||
return hadOld, fmt.Errorf("ensure parent of %s: %w", live, mkErr)
|
||||
}
|
||||
if rerr := os.Rename(tmp, live); rerr != nil {
|
||||
if hadOld {
|
||||
_ = os.Rename(old, live) // revert: live is never left missing
|
||||
}
|
||||
return hadOld, fmt.Errorf("promote restored data into %s: %w", live, rerr)
|
||||
}
|
||||
return hadOld, nil
|
||||
}
|
||||
|
||||
// rollbackSwaps reverts completed swaps in reverse order: drop the restored
|
||||
// live dir and move the preserved original back. Best-effort — each step is
|
||||
// logged by the caller; rollback must attempt every volume regardless.
|
||||
func rollbackSwaps(done []swap) {
|
||||
for i := len(done) - 1; i >= 0; i-- {
|
||||
s := done[i]
|
||||
_ = os.RemoveAll(s.live)
|
||||
if s.hadOld {
|
||||
_ = os.Rename(s.old, s.live)
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,400 @@
|
||||
package volsnap
|
||||
|
||||
import (
|
||||
"context"
|
||||
"encoding/json"
|
||||
"fmt"
|
||||
"log/slog"
|
||||
"os"
|
||||
"path/filepath"
|
||||
"strings"
|
||||
|
||||
"github.com/google/uuid"
|
||||
|
||||
"github.com/alexei/tinyforge/internal/store"
|
||||
)
|
||||
|
||||
// Lifecycle is the deploy-side seam Engine.Restore needs but volsnap must not
|
||||
// import directly (it would couple the snapshot package to the deployer/docker
|
||||
// packages). The composition root supplies an adapter over the Deployer +
|
||||
// Docker client via Engine.SetLifecycle.
|
||||
type Lifecycle interface {
|
||||
// Lock acquires the per-workload deploy lock (C1) and returns the release
|
||||
// func. Held by Restore across stop→swap→redeploy.
|
||||
Lock(workloadID string) func()
|
||||
// StopContainers stops every running container for the workload (C4 quiesce)
|
||||
// and returns the image tag the newest running container was on, so the
|
||||
// redeploy can bring the SAME version back up ("" ⇒ source default tag).
|
||||
StopContainers(ctx context.Context, workloadID string) (runningTag string, err error)
|
||||
// Redeploy re-dispatches the workload's current config WITHOUT re-acquiring
|
||||
// the per-workload lock (the caller holds it). reference pins the image tag.
|
||||
Redeploy(ctx context.Context, w store.Workload, reference string) error
|
||||
}
|
||||
|
||||
// SetLifecycle wires the deploy-side seam. Pass nil to leave restore disabled.
|
||||
func (e *Engine) SetLifecycle(lc Lifecycle) { e.lifecycle = lc }
|
||||
|
||||
// restoreJournal is the on-disk write-ahead record of an in-flight restore.
|
||||
// Written before the first destructive rename and deleted on completion; the
|
||||
// startup RecoverInterruptedRestores sweep replays it after a crash.
|
||||
type restoreJournal struct {
|
||||
SnapshotID string `json:"snapshot_id"`
|
||||
WorkloadID string `json:"workload_id"`
|
||||
Volumes []journalVolume `json:"volumes"`
|
||||
}
|
||||
|
||||
type journalVolume struct {
|
||||
Live string `json:"live"`
|
||||
Old string `json:"old"`
|
||||
Tmp string `json:"tmp"`
|
||||
Swapped bool `json:"swapped"`
|
||||
HadOld bool `json:"had_old"`
|
||||
}
|
||||
|
||||
// staged pairs a resolved volume with its per-restore staging dirs.
|
||||
type staged struct {
|
||||
rv resolvedVol
|
||||
tmp string
|
||||
old string
|
||||
}
|
||||
|
||||
// Restore overwrites the workload's live host-bind volumes with a snapshot's
|
||||
// contents and brings the app back up. It is the single, engine-owned entry
|
||||
// point for the data-loss-sensitive restore flow (image-source workloads only).
|
||||
//
|
||||
// Ordering is deliberate and crash-aware:
|
||||
//
|
||||
// pre-flight (re-resolve all volumes C3, size + per-fs disk check C5) — abort
|
||||
// here touches nothing
|
||||
// → Lock (C1) → re-validate workload → StopContainers (C4 quiesce)
|
||||
// → extract ALL volumes to sibling .tmp staging dirs (reads the source archive
|
||||
// fully BEFORE the next step can prune it; shrinks the later destructive
|
||||
// window to pure renames — R3)
|
||||
// → capture a pre-restore snapshot (durable escape hatch, after quiesce,
|
||||
// before any destructive rename — folded suggestion)
|
||||
// → write the restore journal (R3 crash recovery)
|
||||
// → swap each volume atomically (rename live→.old, .tmp→live — C2)
|
||||
// → Redeploy (C4 — image containers are recreated, never reused)
|
||||
// → remove .old + journal, emit audit event
|
||||
//
|
||||
// Engine.Restore holds NO e.mu (R1): per-workload serialization is the
|
||||
// Lifecycle lock; e.Create takes its own e.mu for the pre-restore archive
|
||||
// write, so calling it here cannot self-deadlock.
|
||||
func (e *Engine) Restore(ctx context.Context, snapshotID, workloadID string) error {
|
||||
if e.lifecycle == nil {
|
||||
return fmt.Errorf("restore: lifecycle not configured")
|
||||
}
|
||||
|
||||
snap, err := e.store.GetVolumeSnapshot(snapshotID)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
if snap.WorkloadID != workloadID {
|
||||
return fmt.Errorf("snapshot %s does not belong to workload %s", snapshotID, workloadID)
|
||||
}
|
||||
w, err := e.store.GetWorkloadByID(workloadID)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
if w.SourceKind != "image" {
|
||||
return fmt.Errorf("restore is only supported for image-source workloads")
|
||||
}
|
||||
settings, err := e.store.GetSettings()
|
||||
if err != nil {
|
||||
return fmt.Errorf("load settings: %w", err)
|
||||
}
|
||||
|
||||
manifest, err := parseManifest(snap)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
resolved, err := preflightResolve(e.store, w, settings, manifest) // C3 all-or-nothing
|
||||
if err != nil {
|
||||
return fmt.Errorf("pre-flight: %w", err)
|
||||
}
|
||||
archivePath, err := e.FilePath(snap)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
perIndex, _, err := archiveUncompressedSize(archivePath, maxRestoreUncompressedBytes)
|
||||
if err != nil {
|
||||
return fmt.Errorf("size snapshot: %w", err)
|
||||
}
|
||||
if err := checkDiskSpace(resolved, perIndex); err != nil { // C5
|
||||
return err
|
||||
}
|
||||
|
||||
// ── past pre-flight: take the per-workload lock and quiesce ──────────────
|
||||
unlock := e.lifecycle.Lock(workloadID)
|
||||
defer unlock()
|
||||
|
||||
// A teardown may have won the lock and deleted the workload while we waited.
|
||||
if _, err := e.store.GetWorkloadByID(workloadID); err != nil {
|
||||
return fmt.Errorf("workload disappeared before restore: %w", err)
|
||||
}
|
||||
|
||||
tag, err := e.lifecycle.StopContainers(ctx, workloadID) // C4 stop
|
||||
if err != nil {
|
||||
return fmt.Errorf("stop containers: %w", err)
|
||||
}
|
||||
|
||||
// Extract every volume to its staging dir FIRST. This reads the source
|
||||
// archive fully before the pre-restore capture below can prune it, and
|
||||
// leaves only pure renames for the destructive phase (R3).
|
||||
token := uuid.New().String()[:8]
|
||||
stagedVols := make([]staged, 0, len(resolved))
|
||||
for _, rv := range resolved {
|
||||
tmp, old := stagingDirs(rv.LivePath, token, rv.Index)
|
||||
if _, exErr := safeExtractIndex(archivePath, rv.Index, tmp, maxRestoreUncompressedBytes); exErr != nil {
|
||||
cleanupStaging(stagedVols)
|
||||
_ = os.RemoveAll(tmp)
|
||||
// Nothing swapped yet — bring the app back up on its original data.
|
||||
e.redeployAfterAbort(ctx, w, tag)
|
||||
return fmt.Errorf("extract volume %q: %w", rv.Target, exErr)
|
||||
}
|
||||
stagedVols = append(stagedVols, staged{rv: rv, tmp: tmp, old: old})
|
||||
}
|
||||
|
||||
// Durable pre-restore snapshot (escape hatch). Quiesced (after stop), and
|
||||
// the source archive is already fully extracted so a prune here is harmless.
|
||||
// Best-effort, matching the DB-restore precedent: a failure is logged but
|
||||
// does not abort — the .old dirs + journal are the in-operation safety net.
|
||||
if _, err := e.Create(w, settings, "pre-restore"); err != nil {
|
||||
slog.Warn("restore: pre-restore snapshot failed (continuing)",
|
||||
"workload", workloadID, "error", err)
|
||||
}
|
||||
|
||||
// Journal before the first destructive rename so a crash can be recovered.
|
||||
jr := restoreJournal{SnapshotID: snapshotID, WorkloadID: workloadID}
|
||||
for _, sv := range stagedVols {
|
||||
jr.Volumes = append(jr.Volumes, journalVolume{Live: sv.rv.LivePath, Old: sv.old, Tmp: sv.tmp})
|
||||
}
|
||||
if err := e.writeJournal(jr); err != nil {
|
||||
cleanupStaging(stagedVols)
|
||||
e.redeployAfterAbort(ctx, w, tag)
|
||||
return fmt.Errorf("write restore journal: %w", err)
|
||||
}
|
||||
|
||||
// ── destructive phase: pure atomic renames ──────────────────────────────
|
||||
done := make([]swap, 0, len(stagedVols))
|
||||
for i, sv := range stagedVols {
|
||||
hadOld, swErr := swapVolumeDir(sv.rv.LivePath, sv.tmp, sv.old)
|
||||
if swErr != nil {
|
||||
rollbackSwaps(done) // restore already-swapped volumes
|
||||
cleanupStagingFrom(stagedVols, i) // drop remaining un-swapped tmp/old
|
||||
e.removeJournal(workloadID)
|
||||
e.redeployAfterAbort(ctx, w, tag)
|
||||
return fmt.Errorf("swap volume %q: %w", sv.rv.Target, swErr)
|
||||
}
|
||||
done = append(done, swap{live: sv.rv.LivePath, old: sv.old, tmp: sv.tmp, hadOld: hadOld})
|
||||
jr.Volumes[i].Swapped = true
|
||||
jr.Volumes[i].HadOld = hadOld
|
||||
_ = e.writeJournal(jr) // progress checkpoint (best-effort)
|
||||
}
|
||||
|
||||
// Bring the app back up against the restored data (C4 — recreate, redeploy).
|
||||
if err := e.lifecycle.Redeploy(ctx, w, tag); err != nil {
|
||||
// The data IS restored; only the app failed to come back. Do NOT roll
|
||||
// back the volumes — surface the redeploy error so the operator retries
|
||||
// a deploy. Clean up the .old set-asides and the journal.
|
||||
cleanupOld(done)
|
||||
e.removeJournal(workloadID)
|
||||
return fmt.Errorf("redeploy after restore: %w", err)
|
||||
}
|
||||
|
||||
cleanupOld(done)
|
||||
e.removeJournal(workloadID)
|
||||
e.emitRestoreEvent(workloadID, snapshotID, len(done))
|
||||
slog.Info("volume snapshot restored", "workload", workloadID, "snapshot", snapshotID, "volumes", len(done))
|
||||
return nil
|
||||
}
|
||||
|
||||
// redeployAfterAbort re-dispatches after an aborted restore so a stopped app
|
||||
// does not stay down. Best-effort: the error is logged, not returned (the
|
||||
// restore failure is the primary error the caller surfaces).
|
||||
func (e *Engine) redeployAfterAbort(ctx context.Context, w store.Workload, tag string) {
|
||||
if err := e.lifecycle.Redeploy(ctx, w, tag); err != nil {
|
||||
slog.Warn("restore: redeploy after abort failed", "workload", w.ID, "error", err)
|
||||
}
|
||||
}
|
||||
|
||||
// RecoverInterruptedRestores replays restore journals left by a crash mid-
|
||||
// restore, mirroring CleanOrphans (run once at startup, before serving). For
|
||||
// each volume: a completed swap keeps the restored live dir and drops the set-
|
||||
// aside original; an incomplete swap that left live missing is reverted from
|
||||
// .old; stray staging dirs are removed. Returns the number of journals handled.
|
||||
func (e *Engine) RecoverInterruptedRestores() (int, error) {
|
||||
e.mu.Lock()
|
||||
defer e.mu.Unlock()
|
||||
|
||||
entries, err := os.ReadDir(e.snapDir)
|
||||
if err != nil {
|
||||
return 0, fmt.Errorf("read snapshot dir: %w", err)
|
||||
}
|
||||
recovered := 0
|
||||
for _, ent := range entries {
|
||||
name := ent.Name()
|
||||
if ent.IsDir() || !strings.HasPrefix(name, "restore-") || !strings.HasSuffix(name, ".json") {
|
||||
continue
|
||||
}
|
||||
path := filepath.Join(e.snapDir, name)
|
||||
data, rerr := os.ReadFile(path)
|
||||
if rerr != nil {
|
||||
slog.Warn("restore recovery: read journal", "file", name, "error", rerr)
|
||||
continue
|
||||
}
|
||||
var jr restoreJournal
|
||||
if jerr := json.Unmarshal(data, &jr); jerr != nil {
|
||||
slog.Warn("restore recovery: parse journal", "file", name, "error", jerr)
|
||||
continue
|
||||
}
|
||||
slog.Warn("restore recovery: replaying interrupted restore",
|
||||
"workload", jr.WorkloadID, "snapshot", jr.SnapshotID, "volumes", len(jr.Volumes))
|
||||
for _, v := range jr.Volumes {
|
||||
recoverVolume(v)
|
||||
}
|
||||
if rmErr := os.Remove(path); rmErr != nil {
|
||||
slog.Warn("restore recovery: remove journal", "file", name, "error", rmErr)
|
||||
}
|
||||
recovered++
|
||||
}
|
||||
return recovered, nil
|
||||
}
|
||||
|
||||
// recoverVolume reconciles a single volume's on-disk state from its journal
|
||||
// entry after a crash. Each branch leaves the live dir intact (either restored
|
||||
// or original) and removes staging leftovers.
|
||||
func recoverVolume(v journalVolume) {
|
||||
if v.Swapped {
|
||||
// Swap completed: live already holds restored data. Drop the set-aside.
|
||||
_ = os.RemoveAll(v.Old)
|
||||
_ = os.RemoveAll(v.Tmp)
|
||||
return
|
||||
}
|
||||
if _, err := os.Lstat(v.Live); os.IsNotExist(err) {
|
||||
if _, oerr := os.Lstat(v.Old); oerr == nil {
|
||||
// Crashed mid-rename (live→old done, tmp→live not): revert.
|
||||
_ = os.Rename(v.Old, v.Live)
|
||||
}
|
||||
} else {
|
||||
// live is intact (original). Any .old is a dangling partial copy.
|
||||
_ = os.RemoveAll(v.Old)
|
||||
}
|
||||
_ = os.RemoveAll(v.Tmp)
|
||||
}
|
||||
|
||||
// ── journal + cleanup helpers ───────────────────────────────────────────────
|
||||
|
||||
func (e *Engine) journalPath(workloadID string) string {
|
||||
// workloadID is a server-generated id (loaded from the DB before we get
|
||||
// here). filepath.Base defends against any separator sneaking into the name.
|
||||
return filepath.Join(e.snapDir, "restore-"+filepath.Base(workloadID)+".json")
|
||||
}
|
||||
|
||||
func (e *Engine) writeJournal(jr restoreJournal) error {
|
||||
data, err := json.Marshal(jr)
|
||||
if err != nil {
|
||||
return fmt.Errorf("encode journal: %w", err)
|
||||
}
|
||||
// Write atomically (tmp + rename): a torn journal would silently disable the
|
||||
// recovery sweep (RecoverInterruptedRestores skips unparseable journals), so
|
||||
// a crash mid-write must never leave a half-written WAL on disk. The .tmp
|
||||
// suffix is ignored by the recovery scan (it matches *.json only).
|
||||
final := e.journalPath(jr.WorkloadID)
|
||||
tmp := final + ".tmp"
|
||||
if err := os.WriteFile(tmp, data, 0o600); err != nil {
|
||||
return fmt.Errorf("write journal: %w", err)
|
||||
}
|
||||
if err := os.Rename(tmp, final); err != nil {
|
||||
_ = os.Remove(tmp)
|
||||
return fmt.Errorf("commit journal: %w", err)
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
func (e *Engine) removeJournal(workloadID string) {
|
||||
if err := os.Remove(e.journalPath(workloadID)); err != nil && !os.IsNotExist(err) {
|
||||
slog.Warn("restore: remove journal", "workload", workloadID, "error", err)
|
||||
}
|
||||
}
|
||||
|
||||
func (e *Engine) emitRestoreEvent(workloadID, snapshotID string, volumes int) {
|
||||
meta, _ := json.Marshal(map[string]any{"snapshot_id": snapshotID, "volumes": volumes})
|
||||
if _, err := e.store.InsertEvent(store.EventLog{
|
||||
Source: "volsnap",
|
||||
WorkloadID: workloadID,
|
||||
Severity: "info",
|
||||
Message: "volume snapshot restored",
|
||||
Metadata: string(meta),
|
||||
}); err != nil {
|
||||
slog.Warn("restore: record event", "workload", workloadID, "error", err)
|
||||
}
|
||||
}
|
||||
|
||||
// cleanupStaging removes the tmp + old staging dirs for every staged volume
|
||||
// (used when aborting before the swap phase).
|
||||
func cleanupStaging(sv []staged) {
|
||||
for _, s := range sv {
|
||||
_ = os.RemoveAll(s.tmp)
|
||||
_ = os.RemoveAll(s.old)
|
||||
}
|
||||
}
|
||||
|
||||
// cleanupStagingFrom removes staging dirs from index `from` onward (the volumes
|
||||
// not yet swapped when a swap failed).
|
||||
func cleanupStagingFrom(sv []staged, from int) {
|
||||
for i := from; i < len(sv); i++ {
|
||||
_ = os.RemoveAll(sv[i].tmp)
|
||||
_ = os.RemoveAll(sv[i].old)
|
||||
}
|
||||
}
|
||||
|
||||
// cleanupOld removes the .old set-aside dirs after a successful (or data-
|
||||
// committed) restore to reclaim disk; the pre-restore snapshot is the durable
|
||||
// rollback target.
|
||||
func cleanupOld(done []swap) {
|
||||
for _, s := range done {
|
||||
_ = os.RemoveAll(s.old)
|
||||
}
|
||||
}
|
||||
|
||||
// checkDiskSpace verifies each target filesystem has room for the volumes that
|
||||
// will be staged on it (C5). Peak usage co-locates the live copy (renamed
|
||||
// aside, no new space) and the extracted copy (new space ≈ uncompressed size),
|
||||
// so the new allocation per filesystem is the sum of its volumes' extracted
|
||||
// sizes plus headroom. The estimate is a lower bound (see archiveUncompressedSize);
|
||||
// a mid-extract ENOSPC is still caught and rolled back.
|
||||
func checkDiskSpace(resolved []resolvedVol, perIndex map[int]int64) error {
|
||||
needByParent := map[string]int64{}
|
||||
for _, rv := range resolved {
|
||||
needByParent[filepath.Dir(rv.LivePath)] += perIndex[rv.Index]
|
||||
}
|
||||
for parent, need := range needByParent {
|
||||
probe := firstExistingAncestor(parent)
|
||||
free, err := freeDiskBytes(probe)
|
||||
if err != nil {
|
||||
return fmt.Errorf("check disk space at %s: %w", probe, err)
|
||||
}
|
||||
if int64(free) < need+diskFreeHeadroomBytes {
|
||||
return fmt.Errorf("insufficient disk space at %s: need ~%d bytes, have %d",
|
||||
parent, need+diskFreeHeadroomBytes, free)
|
||||
}
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
// firstExistingAncestor walks up p until it finds a path that exists, so the
|
||||
// free-space probe has a real filesystem to stat even when the volume dir (or
|
||||
// its parent) hasn't been created yet.
|
||||
func firstExistingAncestor(p string) string {
|
||||
for {
|
||||
if _, err := os.Stat(p); err == nil {
|
||||
return p
|
||||
}
|
||||
parent := filepath.Dir(p)
|
||||
if parent == p {
|
||||
return p
|
||||
}
|
||||
p = parent
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,345 @@
|
||||
package volsnap
|
||||
|
||||
import (
|
||||
"archive/tar"
|
||||
"context"
|
||||
"errors"
|
||||
"os"
|
||||
"path/filepath"
|
||||
"strings"
|
||||
"sync"
|
||||
"testing"
|
||||
|
||||
"github.com/alexei/tinyforge/internal/store"
|
||||
"github.com/alexei/tinyforge/internal/volume"
|
||||
)
|
||||
|
||||
// fakeLifecycle records the order of deploy-side calls and lets tests inject
|
||||
// failures, without a real deployer/docker.
|
||||
type fakeLifecycle struct {
|
||||
mu sync.Mutex
|
||||
calls []string
|
||||
tag string
|
||||
stopErr error
|
||||
redeployErr error
|
||||
redeployRef string
|
||||
}
|
||||
|
||||
func (f *fakeLifecycle) rec(s string) {
|
||||
f.mu.Lock()
|
||||
f.calls = append(f.calls, s)
|
||||
f.mu.Unlock()
|
||||
}
|
||||
func (f *fakeLifecycle) Lock(string) func() { f.rec("lock"); return func() { f.rec("unlock") } }
|
||||
func (f *fakeLifecycle) StopContainers(context.Context, string) (string, error) {
|
||||
f.rec("stop")
|
||||
return f.tag, f.stopErr
|
||||
}
|
||||
func (f *fakeLifecycle) Redeploy(_ context.Context, _ store.Workload, ref string) error {
|
||||
f.rec("redeploy:" + ref)
|
||||
f.redeployRef = ref
|
||||
return f.redeployErr
|
||||
}
|
||||
func (f *fakeLifecycle) saw(s string) bool {
|
||||
f.mu.Lock()
|
||||
defer f.mu.Unlock()
|
||||
for _, c := range f.calls {
|
||||
if c == s {
|
||||
return true
|
||||
}
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
func newRestoreEngine(t *testing.T) (*Engine, *store.Store, string) {
|
||||
t.Helper()
|
||||
st, err := store.New(":memory:")
|
||||
if err != nil {
|
||||
t.Fatalf("store: %v", err)
|
||||
}
|
||||
t.Cleanup(func() { st.Close() })
|
||||
base := t.TempDir()
|
||||
s, _ := st.GetSettings()
|
||||
s.BaseVolumePath = base
|
||||
if err := st.UpdateSettings(s); err != nil {
|
||||
t.Fatalf("settings: %v", err)
|
||||
}
|
||||
eng, err := New(st, t.TempDir())
|
||||
if err != nil {
|
||||
t.Fatalf("engine: %v", err)
|
||||
}
|
||||
return eng, st, base
|
||||
}
|
||||
|
||||
// seedImageWorkload creates an image workload with one project-scope volume and
|
||||
// returns it plus the resolved live host dir.
|
||||
func seedImageWorkload(t *testing.T, st *store.Store) (store.Workload, string) {
|
||||
t.Helper()
|
||||
w, err := st.CreateWorkload(store.Workload{
|
||||
Name: "data-app",
|
||||
Kind: "project",
|
||||
SourceKind: "image",
|
||||
SourceConfig: `{"image":"reg/app","port":80,"volumes":[{"source":"data","target":"/data","scope":"project"}]}`,
|
||||
})
|
||||
if err != nil {
|
||||
t.Fatalf("create workload: %v", err)
|
||||
}
|
||||
settings, _ := st.GetSettings()
|
||||
live, err := volume.ResolveWorkloadPath(
|
||||
store.WorkloadVolume{Source: "data", Target: "/data", Scope: "project"},
|
||||
volume.ResolveWorkloadParams{BasePath: settings.BaseVolumePath, WorkloadID: w.ID, WorkloadName: w.Name},
|
||||
)
|
||||
if err != nil {
|
||||
t.Fatalf("resolve: %v", err)
|
||||
}
|
||||
return w, live
|
||||
}
|
||||
|
||||
func TestEngineRestore_HappyPath(t *testing.T) {
|
||||
eng, st, _ := newRestoreEngine(t)
|
||||
w, live := seedImageWorkload(t, st)
|
||||
mkDirWith(t, live, "orig.txt", "ORIGINAL")
|
||||
|
||||
settings, _ := st.GetSettings()
|
||||
snap, err := eng.Create(w, settings, "base")
|
||||
if err != nil {
|
||||
t.Fatalf("create snapshot: %v", err)
|
||||
}
|
||||
|
||||
// Drift: the live dir now differs from the snapshot.
|
||||
if err := os.WriteFile(filepath.Join(live, "orig.txt"), []byte("CHANGED"), 0o600); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
mkDirWith(t, live, "extra.txt", "NEW") // not in the snapshot
|
||||
|
||||
fake := &fakeLifecycle{tag: "v1.2.3"}
|
||||
eng.SetLifecycle(fake)
|
||||
|
||||
// Uses the REAL eng.Create for the pre-restore capture — if Restore held
|
||||
// e.mu this would deadlock (R1), failing the test instead of production.
|
||||
if err := eng.Restore(context.Background(), snap.ID, w.ID); err != nil {
|
||||
t.Fatalf("restore: %v", err)
|
||||
}
|
||||
|
||||
if got := readIn(t, live, "orig.txt"); got != "ORIGINAL" {
|
||||
t.Errorf("orig.txt = %q, want ORIGINAL (restored)", got)
|
||||
}
|
||||
if _, err := os.Stat(filepath.Join(live, "extra.txt")); !os.IsNotExist(err) {
|
||||
t.Error("extra.txt should be gone — restore replaces the volume dir wholesale")
|
||||
}
|
||||
for _, want := range []string{"lock", "stop", "redeploy:v1.2.3", "unlock"} {
|
||||
if !fake.saw(want) {
|
||||
t.Errorf("expected lifecycle call %q; calls=%v", want, fake.calls)
|
||||
}
|
||||
}
|
||||
if fake.redeployRef != "v1.2.3" {
|
||||
t.Errorf("redeploy reference = %q, want the running tag v1.2.3", fake.redeployRef)
|
||||
}
|
||||
// A durable pre-restore snapshot was captured (base + pre-restore).
|
||||
snaps, _ := eng.List(w.ID)
|
||||
if len(snaps) != 2 {
|
||||
t.Errorf("expected 2 snapshots (base + pre-restore), got %d", len(snaps))
|
||||
}
|
||||
// No journal left behind.
|
||||
assertNoJournal(t, eng)
|
||||
}
|
||||
|
||||
func TestEngineRestore_RedeployFailureKeepsRestoredData(t *testing.T) {
|
||||
eng, st, _ := newRestoreEngine(t)
|
||||
w, live := seedImageWorkload(t, st)
|
||||
mkDirWith(t, live, "orig.txt", "ORIGINAL")
|
||||
settings, _ := st.GetSettings()
|
||||
snap, _ := eng.Create(w, settings, "base")
|
||||
if err := os.WriteFile(filepath.Join(live, "orig.txt"), []byte("CHANGED"), 0o600); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
|
||||
fake := &fakeLifecycle{tag: "v1", redeployErr: errors.New("boom")}
|
||||
eng.SetLifecycle(fake)
|
||||
|
||||
err := eng.Restore(context.Background(), snap.ID, w.ID)
|
||||
if err == nil || !strings.Contains(err.Error(), "redeploy") {
|
||||
t.Fatalf("expected a redeploy error, got %v", err)
|
||||
}
|
||||
// Data is committed despite the redeploy failure — we must NOT roll it back.
|
||||
if got := readIn(t, live, "orig.txt"); got != "ORIGINAL" {
|
||||
t.Errorf("orig.txt = %q, want ORIGINAL (restore committed)", got)
|
||||
}
|
||||
assertNoJournal(t, eng)
|
||||
}
|
||||
|
||||
func TestEngineRestore_PreflightFailDoesNotLockOrStop(t *testing.T) {
|
||||
eng, st, _ := newRestoreEngine(t)
|
||||
w, _ := seedImageWorkload(t, st)
|
||||
// A snapshot whose manifest names an unsupported scope ⇒ pre-flight aborts.
|
||||
bad, err := st.CreateVolumeSnapshot(store.VolumeSnapshot{
|
||||
WorkloadID: w.ID, Filename: "bad.tar.gz",
|
||||
Manifest: `[{"index":0,"target":"/x","scope":"named","source":"x"}]`,
|
||||
})
|
||||
if err != nil {
|
||||
t.Fatalf("seed snapshot: %v", err)
|
||||
}
|
||||
fake := &fakeLifecycle{}
|
||||
eng.SetLifecycle(fake)
|
||||
|
||||
if err := eng.Restore(context.Background(), bad.ID, w.ID); err == nil {
|
||||
t.Fatal("expected pre-flight to abort on an unsupported scope")
|
||||
}
|
||||
if fake.saw("lock") || fake.saw("stop") {
|
||||
t.Errorf("pre-flight abort must happen BEFORE lock/stop; calls=%v", fake.calls)
|
||||
}
|
||||
}
|
||||
|
||||
func TestEngineRestore_NilLifecycle(t *testing.T) {
|
||||
eng, _, _ := newRestoreEngine(t)
|
||||
if err := eng.Restore(context.Background(), "s", "w"); err == nil ||
|
||||
!strings.Contains(err.Error(), "lifecycle") {
|
||||
t.Fatalf("expected a lifecycle-not-configured error, got %v", err)
|
||||
}
|
||||
}
|
||||
|
||||
func TestEngineRestore_WrongWorkload(t *testing.T) {
|
||||
eng, st, _ := newRestoreEngine(t)
|
||||
w, live := seedImageWorkload(t, st)
|
||||
mkDirWith(t, live, "f.txt", "x")
|
||||
settings, _ := st.GetSettings()
|
||||
snap, _ := eng.Create(w, settings, "base")
|
||||
fake := &fakeLifecycle{}
|
||||
eng.SetLifecycle(fake)
|
||||
|
||||
if err := eng.Restore(context.Background(), snap.ID, "some-other-workload"); err == nil {
|
||||
t.Fatal("expected cross-workload restore to be rejected")
|
||||
}
|
||||
if fake.saw("lock") {
|
||||
t.Error("must reject before taking the lock")
|
||||
}
|
||||
}
|
||||
|
||||
func TestEngineRestore_ExtractFailureAbortsAfterLock(t *testing.T) {
|
||||
eng, st, _ := newRestoreEngine(t)
|
||||
// The workload must CURRENTLY declare both targets so pre-flight passes and
|
||||
// the failure happens during extraction (post-lock), not pre-flight.
|
||||
w, err := st.CreateWorkload(store.Workload{
|
||||
Name: "two-vol", Kind: "project", SourceKind: "image",
|
||||
SourceConfig: `{"image":"x","port":80,"volumes":[` +
|
||||
`{"source":"data","target":"/data","scope":"project"},` +
|
||||
`{"source":"other","target":"/other","scope":"project"}]}`,
|
||||
})
|
||||
if err != nil {
|
||||
t.Fatalf("create workload: %v", err)
|
||||
}
|
||||
|
||||
// Hand-build a 2-volume archive where volume 1 carries a symlink entry the
|
||||
// untrusted extractor rejects — forcing a post-lock extract failure after
|
||||
// volume 0 has already been staged.
|
||||
arc := buildTarGz(t, []tentry{
|
||||
{name: "0/f.txt", typeflag: tar.TypeReg, body: "x"},
|
||||
{name: "1/evil", typeflag: tar.TypeSymlink, linkname: "/etc/passwd"},
|
||||
})
|
||||
data, err := os.ReadFile(arc)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
fname := "extract-fail.tar.gz"
|
||||
if err := os.WriteFile(filepath.Join(eng.snapDir, fname), data, 0o600); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
snap, err := st.CreateVolumeSnapshot(store.VolumeSnapshot{
|
||||
WorkloadID: w.ID, Filename: fname,
|
||||
Manifest: `[{"index":0,"target":"/data","scope":"project","source":"data"},` +
|
||||
`{"index":1,"target":"/other","scope":"project","source":"other"}]`,
|
||||
})
|
||||
if err != nil {
|
||||
t.Fatalf("seed snapshot: %v", err)
|
||||
}
|
||||
|
||||
fake := &fakeLifecycle{tag: "v1"}
|
||||
eng.SetLifecycle(fake)
|
||||
|
||||
if err := eng.Restore(context.Background(), snap.ID, w.ID); err == nil {
|
||||
t.Fatal("expected extract failure to abort the restore")
|
||||
}
|
||||
// Post-lock abort: it stopped, then brought the app back (no swaps happened).
|
||||
if !fake.saw("lock") || !fake.saw("stop") || !fake.saw("redeploy:v1") {
|
||||
t.Errorf("expected lock+stop+redeploy after a post-lock abort; calls=%v", fake.calls)
|
||||
}
|
||||
// No staging or journal left behind.
|
||||
assertNoJournal(t, eng)
|
||||
entries, _ := os.ReadDir(eng.snapDir)
|
||||
for _, e := range entries {
|
||||
if strings.Contains(e.Name(), ".tf-restore-") {
|
||||
t.Errorf("leftover staging dir: %s", e.Name())
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func TestRecoverInterruptedRestores(t *testing.T) {
|
||||
eng, _, _ := newRestoreEngine(t)
|
||||
root := t.TempDir()
|
||||
|
||||
// A: swap completed — keep restored live, drop old.
|
||||
liveA := filepath.Join(root, "A")
|
||||
oldA := filepath.Join(root, ".A.old")
|
||||
mkDirWith(t, liveA, "f", "RESTORED-A")
|
||||
mkDirWith(t, oldA, "f", "ORIGINAL-A")
|
||||
// B: not swapped, live present — keep original, drop tmp.
|
||||
liveB := filepath.Join(root, "B")
|
||||
tmpB := filepath.Join(root, ".B.tmp")
|
||||
mkDirWith(t, liveB, "f", "ORIGINAL-B")
|
||||
mkDirWith(t, tmpB, "f", "STAGED-B")
|
||||
// C: crashed mid-rename — live missing, old present — revert from old.
|
||||
liveC := filepath.Join(root, "C")
|
||||
oldC := filepath.Join(root, ".C.old")
|
||||
tmpC := filepath.Join(root, ".C.tmp")
|
||||
mkDirWith(t, oldC, "f", "ORIGINAL-C")
|
||||
mkDirWith(t, tmpC, "f", "STAGED-C")
|
||||
|
||||
jr := restoreJournal{SnapshotID: "snap", WorkloadID: "wl-recover", Volumes: []journalVolume{
|
||||
{Live: liveA, Old: oldA, Swapped: true, HadOld: true},
|
||||
{Live: liveB, Tmp: tmpB, Swapped: false},
|
||||
{Live: liveC, Old: oldC, Tmp: tmpC, Swapped: false},
|
||||
}}
|
||||
if err := eng.writeJournal(jr); err != nil {
|
||||
t.Fatalf("write journal: %v", err)
|
||||
}
|
||||
|
||||
n, err := eng.RecoverInterruptedRestores()
|
||||
if err != nil {
|
||||
t.Fatalf("recover: %v", err)
|
||||
}
|
||||
if n != 1 {
|
||||
t.Fatalf("recovered %d journals, want 1", n)
|
||||
}
|
||||
if got := readIn(t, liveA, "f"); got != "RESTORED-A" {
|
||||
t.Errorf("A live = %q, want RESTORED-A (swap kept)", got)
|
||||
}
|
||||
if _, err := os.Stat(oldA); !os.IsNotExist(err) {
|
||||
t.Error("A old should be removed")
|
||||
}
|
||||
if got := readIn(t, liveB, "f"); got != "ORIGINAL-B" {
|
||||
t.Errorf("B live = %q, want ORIGINAL-B (untouched)", got)
|
||||
}
|
||||
if _, err := os.Stat(tmpB); !os.IsNotExist(err) {
|
||||
t.Error("B tmp should be removed")
|
||||
}
|
||||
if got := readIn(t, liveC, "f"); got != "ORIGINAL-C" {
|
||||
t.Errorf("C live = %q, want ORIGINAL-C (reverted from old)", got)
|
||||
}
|
||||
if _, err := os.Stat(tmpC); !os.IsNotExist(err) {
|
||||
t.Error("C tmp should be removed")
|
||||
}
|
||||
assertNoJournal(t, eng)
|
||||
}
|
||||
|
||||
func assertNoJournal(t *testing.T, eng *Engine) {
|
||||
t.Helper()
|
||||
entries, err := os.ReadDir(eng.snapDir)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
for _, e := range entries {
|
||||
if strings.HasPrefix(e.Name(), "restore-") && strings.HasSuffix(e.Name(), ".json") {
|
||||
t.Errorf("leftover restore journal: %s", e.Name())
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,270 @@
|
||||
package volsnap
|
||||
|
||||
import (
|
||||
"os"
|
||||
"path/filepath"
|
||||
"testing"
|
||||
|
||||
"github.com/alexei/tinyforge/internal/store"
|
||||
)
|
||||
|
||||
func mkDirWith(t *testing.T, dir, fname, content string) {
|
||||
t.Helper()
|
||||
if err := os.MkdirAll(dir, 0o700); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if err := os.WriteFile(filepath.Join(dir, fname), []byte(content), 0o600); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
}
|
||||
|
||||
func readIn(t *testing.T, dir, fname string) string {
|
||||
t.Helper()
|
||||
b, err := os.ReadFile(filepath.Join(dir, fname))
|
||||
if err != nil {
|
||||
t.Fatalf("read %s/%s: %v", dir, fname, err)
|
||||
}
|
||||
return string(b)
|
||||
}
|
||||
|
||||
func TestSwapVolumeDir_ReplacesAndPreservesOld(t *testing.T) {
|
||||
root := t.TempDir()
|
||||
live := filepath.Join(root, "data")
|
||||
tmp := filepath.Join(root, ".data.tmp")
|
||||
old := filepath.Join(root, ".data.old")
|
||||
mkDirWith(t, live, "f.txt", "ORIGINAL")
|
||||
mkDirWith(t, tmp, "f.txt", "RESTORED")
|
||||
|
||||
hadOld, err := swapVolumeDir(live, tmp, old)
|
||||
if err != nil {
|
||||
t.Fatalf("swap: %v", err)
|
||||
}
|
||||
if !hadOld {
|
||||
t.Error("hadOld should be true when a live dir existed")
|
||||
}
|
||||
if got := readIn(t, live, "f.txt"); got != "RESTORED" {
|
||||
t.Errorf("live = %q, want RESTORED", got)
|
||||
}
|
||||
if got := readIn(t, old, "f.txt"); got != "ORIGINAL" {
|
||||
t.Errorf("old = %q, want ORIGINAL (prior live preserved)", got)
|
||||
}
|
||||
}
|
||||
|
||||
func TestSwapVolumeDir_MissingLive(t *testing.T) {
|
||||
root := t.TempDir()
|
||||
live := filepath.Join(root, "data") // does not exist
|
||||
tmp := filepath.Join(root, ".data.tmp")
|
||||
old := filepath.Join(root, ".data.old")
|
||||
mkDirWith(t, tmp, "f.txt", "RESTORED")
|
||||
|
||||
hadOld, err := swapVolumeDir(live, tmp, old)
|
||||
if err != nil {
|
||||
t.Fatalf("swap: %v", err)
|
||||
}
|
||||
if hadOld {
|
||||
t.Error("hadOld should be false when no live dir existed")
|
||||
}
|
||||
if got := readIn(t, live, "f.txt"); got != "RESTORED" {
|
||||
t.Errorf("live = %q, want RESTORED", got)
|
||||
}
|
||||
}
|
||||
|
||||
func TestSwapVolumeDir_RevertsOnSecondRenameFailure(t *testing.T) {
|
||||
// The data-loss-critical path: the live→old rename succeeds, then tmp→live
|
||||
// fails (here: tmp is absent). swapVolumeDir MUST self-revert old→live so
|
||||
// the live dir is never left missing, and old must not be left dangling.
|
||||
root := t.TempDir()
|
||||
live := filepath.Join(root, "data")
|
||||
old := filepath.Join(root, ".data.old")
|
||||
mkDirWith(t, live, "f.txt", "ORIGINAL")
|
||||
|
||||
hadOld, err := swapVolumeDir(live, filepath.Join(root, ".data.tmp" /* absent */), old)
|
||||
if err == nil {
|
||||
t.Fatal("expected swap to fail when tmp is absent")
|
||||
}
|
||||
if got := readIn(t, live, "f.txt"); got != "ORIGINAL" {
|
||||
t.Errorf("live = %q, want ORIGINAL restored by self-revert", got)
|
||||
}
|
||||
if _, statErr := os.Stat(old); !os.IsNotExist(statErr) {
|
||||
t.Errorf("old dir should have been renamed back to live, not left dangling")
|
||||
}
|
||||
_ = hadOld
|
||||
}
|
||||
|
||||
func TestStagingDirs_SameParentAsLive(t *testing.T) {
|
||||
// R2 invariant: tmp and old must be siblings of the live dir's parent so
|
||||
// every rename in the swap is intra-filesystem (atomic).
|
||||
live := filepath.FromSlash("/srv/data/postgres")
|
||||
tmp, old := stagingDirs(live, "tok", 3)
|
||||
wantParent := filepath.Dir(live)
|
||||
if filepath.Dir(tmp) != wantParent || filepath.Dir(old) != wantParent {
|
||||
t.Errorf("staging dirs not siblings of live's parent: tmp=%s old=%s parent=%s", tmp, old, wantParent)
|
||||
}
|
||||
if tmp == old {
|
||||
t.Error("tmp and old must be distinct paths")
|
||||
}
|
||||
}
|
||||
|
||||
func TestRollbackSwaps_RestoresOriginals(t *testing.T) {
|
||||
root := t.TempDir()
|
||||
var done []swap
|
||||
for _, name := range []string{"vol0", "vol1"} {
|
||||
live := filepath.Join(root, name)
|
||||
tmp := filepath.Join(root, "."+name+".tmp")
|
||||
old := filepath.Join(root, "."+name+".old")
|
||||
mkDirWith(t, live, "f.txt", "ORIGINAL-"+name)
|
||||
mkDirWith(t, tmp, "f.txt", "RESTORED-"+name)
|
||||
hadOld, err := swapVolumeDir(live, tmp, old)
|
||||
if err != nil {
|
||||
t.Fatalf("swap %s: %v", name, err)
|
||||
}
|
||||
done = append(done, swap{live: live, old: old, tmp: tmp, hadOld: hadOld})
|
||||
}
|
||||
// Both are now RESTORED; rolling back must return both to ORIGINAL.
|
||||
rollbackSwaps(done)
|
||||
for _, name := range []string{"vol0", "vol1"} {
|
||||
if got := readIn(t, filepath.Join(root, name), "f.txt"); got != "ORIGINAL-"+name {
|
||||
t.Errorf("%s after rollback = %q, want ORIGINAL-%s", name, got, name)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func TestRollbackSwaps_PartialLeavesUnswappedIntact(t *testing.T) {
|
||||
root := t.TempDir()
|
||||
// vol0 swaps successfully; vol1 "fails" (its tmp is absent) so only vol0 is
|
||||
// recorded in done. Rollback of vol0 must restore the original.
|
||||
live0 := filepath.Join(root, "vol0")
|
||||
tmp0 := filepath.Join(root, ".vol0.tmp")
|
||||
old0 := filepath.Join(root, ".vol0.old")
|
||||
mkDirWith(t, live0, "f.txt", "ORIGINAL-0")
|
||||
mkDirWith(t, tmp0, "f.txt", "RESTORED-0")
|
||||
hadOld, err := swapVolumeDir(live0, tmp0, old0)
|
||||
if err != nil {
|
||||
t.Fatalf("swap vol0: %v", err)
|
||||
}
|
||||
|
||||
live1 := filepath.Join(root, "vol1")
|
||||
mkDirWith(t, live1, "f.txt", "ORIGINAL-1")
|
||||
if _, err := swapVolumeDir(live1, filepath.Join(root, ".vol1.tmp" /* absent */), filepath.Join(root, ".vol1.old")); err == nil {
|
||||
t.Fatal("expected vol1 swap to fail (tmp absent)")
|
||||
}
|
||||
|
||||
rollbackSwaps([]swap{{live: live0, old: old0, tmp: tmp0, hadOld: hadOld}})
|
||||
if got := readIn(t, live0, "f.txt"); got != "ORIGINAL-0" {
|
||||
t.Errorf("vol0 after rollback = %q, want ORIGINAL-0", got)
|
||||
}
|
||||
// vol1 was never swapped; its original must be untouched.
|
||||
if got := readIn(t, live1, "f.txt"); got != "ORIGINAL-1" {
|
||||
t.Errorf("vol1 = %q, want ORIGINAL-1 (never swapped)", got)
|
||||
}
|
||||
}
|
||||
|
||||
func TestPreflightResolve_AllOrNothing(t *testing.T) {
|
||||
eng, st, base := newRestoreEngine(t)
|
||||
_ = eng
|
||||
w, err := st.CreateWorkload(store.Workload{
|
||||
Name: "app", Kind: "project", SourceKind: "image",
|
||||
SourceConfig: `{"image":"x","port":80,"volumes":[` +
|
||||
`{"source":"data","target":"/data","scope":"project"},` +
|
||||
`{"source":"var","target":"/var","scope":"stage"}]}`,
|
||||
})
|
||||
if err != nil {
|
||||
t.Fatalf("create workload: %v", err)
|
||||
}
|
||||
settings, _ := st.GetSettings()
|
||||
|
||||
// A snapshotted target no longer declared by the workload ⇒ whole abort (C3).
|
||||
if _, err := preflightResolve(st, w, settings, []SnapshotVolume{
|
||||
{Index: 0, Target: "/data", Scope: "project", Source: "data"},
|
||||
{Index: 1, Target: "/gone", Scope: "project", Source: "gone"},
|
||||
}); err == nil {
|
||||
t.Fatal("expected all-or-nothing abort when a target is no longer declared")
|
||||
}
|
||||
|
||||
// All declared ⇒ resolves every volume under BaseVolumePath.
|
||||
resolved, err := preflightResolve(st, w, settings, []SnapshotVolume{
|
||||
{Index: 0, Target: "/data", Scope: "project", Source: "data"},
|
||||
{Index: 1, Target: "/var", Scope: "stage", Source: "var"},
|
||||
})
|
||||
if err != nil {
|
||||
t.Fatalf("preflight: %v", err)
|
||||
}
|
||||
if len(resolved) != 2 {
|
||||
t.Fatalf("resolved %d volumes, want 2", len(resolved))
|
||||
}
|
||||
for _, rv := range resolved {
|
||||
if ok, _ := pathWithinBase(base, rv.LivePath); !ok {
|
||||
t.Errorf("volume %q resolved to %q, outside base %q", rv.Target, rv.LivePath, base)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// TestPreflightResolve_IgnoresManifestSource is the regression guard for the
|
||||
// security fix: a tampered manifest whose Source tries to escape (../../etc)
|
||||
// must NOT redirect the swap target — the host path is re-derived from the
|
||||
// workload's CURRENT trusted config (Source "data"), staying under the base.
|
||||
func TestPreflightResolve_IgnoresManifestSource(t *testing.T) {
|
||||
_, st, base := newRestoreEngine(t)
|
||||
w, err := st.CreateWorkload(store.Workload{
|
||||
Name: "app", Kind: "project", SourceKind: "image",
|
||||
SourceConfig: `{"image":"x","port":80,"volumes":[{"source":"data","target":"/data","scope":"project"}]}`,
|
||||
})
|
||||
if err != nil {
|
||||
t.Fatalf("create workload: %v", err)
|
||||
}
|
||||
settings, _ := st.GetSettings()
|
||||
|
||||
resolved, err := preflightResolve(st, w, settings, []SnapshotVolume{
|
||||
{Index: 0, Target: "/data", Scope: "project", Source: "../../../../etc"},
|
||||
})
|
||||
if err != nil {
|
||||
t.Fatalf("preflight: %v", err)
|
||||
}
|
||||
if len(resolved) != 1 {
|
||||
t.Fatalf("resolved %d, want 1", len(resolved))
|
||||
}
|
||||
if ok, _ := pathWithinBase(base, resolved[0].LivePath); !ok {
|
||||
t.Errorf("manifest Source escaped containment: resolved %q outside base %q",
|
||||
resolved[0].LivePath, base)
|
||||
}
|
||||
if filepath.Base(resolved[0].LivePath) != "data" {
|
||||
t.Errorf("expected target derived from current config (data), got %q", resolved[0].LivePath)
|
||||
}
|
||||
}
|
||||
|
||||
func TestArchiveUncompressedSize(t *testing.T) {
|
||||
root := t.TempDir()
|
||||
mustWrite(t, filepath.Join(root, "a.txt"), "hello") // 5
|
||||
if err := os.MkdirAll(filepath.Join(root, "sub"), 0o700); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
mustWrite(t, filepath.Join(root, "sub", "b.txt"), "world!") // 6
|
||||
dest := filepath.Join(t.TempDir(), "snap.tar.gz")
|
||||
if _, err := writeArchive(dest, []VolumeRef{{Target: "/d", Scope: "project", Source: "d", HostPath: root}}); err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
|
||||
per, total, err := archiveUncompressedSize(dest, maxRestoreUncompressedBytes)
|
||||
if err != nil {
|
||||
t.Fatalf("size: %v", err)
|
||||
}
|
||||
if total != 11 {
|
||||
t.Errorf("total = %d, want 11", total)
|
||||
}
|
||||
if per[0] != 11 {
|
||||
t.Errorf("perIndex[0] = %d, want 11", per[0])
|
||||
}
|
||||
if _, _, err := archiveUncompressedSize(dest, 4); err == nil {
|
||||
t.Error("expected sizing to abort past the decompression cap")
|
||||
}
|
||||
}
|
||||
|
||||
func TestFreeDiskBytes(t *testing.T) {
|
||||
n, err := freeDiskBytes(t.TempDir())
|
||||
if err != nil {
|
||||
t.Fatalf("freeDiskBytes: %v", err)
|
||||
}
|
||||
if n == 0 {
|
||||
t.Error("expected non-zero free space on the temp filesystem")
|
||||
}
|
||||
}
|
||||
+32
-18
@@ -80,27 +80,10 @@ func SnapshotableVolumes(st *store.Store, w store.Workload, settings store.Setti
|
||||
return nil, nil, nil
|
||||
}
|
||||
|
||||
byTarget := map[string]store.WorkloadVolume{}
|
||||
|
||||
var cfg scVolumes
|
||||
if w.SourceConfig != "" {
|
||||
// Best-effort: a malformed config simply yields no inline volumes; the
|
||||
// persisted rows below still apply.
|
||||
_ = json.Unmarshal([]byte(w.SourceConfig), &cfg)
|
||||
}
|
||||
for _, v := range cfg.Volumes {
|
||||
if v.Target == "" {
|
||||
continue
|
||||
}
|
||||
byTarget[v.Target] = store.WorkloadVolume{Source: v.Source, Target: v.Target, Scope: v.Scope, Name: v.Name}
|
||||
}
|
||||
persisted, perr := st.ListWorkloadVolumes(w.ID)
|
||||
byTarget, perr := volumesByTarget(st, w)
|
||||
if perr != nil {
|
||||
return nil, nil, perr
|
||||
}
|
||||
for _, p := range persisted {
|
||||
byTarget[p.Target] = store.WorkloadVolume{Source: p.Source, Target: p.Target, Scope: p.Scope, Name: p.Name}
|
||||
}
|
||||
|
||||
params := volume.ResolveWorkloadParams{
|
||||
BasePath: settings.BaseVolumePath,
|
||||
@@ -132,6 +115,37 @@ func SnapshotableVolumes(st *store.Store, w store.Workload, settings store.Setti
|
||||
return refs, skipped, nil
|
||||
}
|
||||
|
||||
// volumesByTarget merges a workload's source_config inline volumes with its
|
||||
// persisted workload_volumes rows (persisted wins on a target conflict), keyed
|
||||
// by container target path. It is the authoritative current volume set, shared
|
||||
// by capture enumeration (SnapshotableVolumes) and restore pre-flight so both
|
||||
// resolve host paths the same way — restore must never trust a snapshot's
|
||||
// persisted manifest to name a host directory.
|
||||
func volumesByTarget(st *store.Store, w store.Workload) (map[string]store.WorkloadVolume, error) {
|
||||
byTarget := map[string]store.WorkloadVolume{}
|
||||
|
||||
var cfg scVolumes
|
||||
if w.SourceConfig != "" {
|
||||
// Best-effort: a malformed config simply yields no inline volumes; the
|
||||
// persisted rows below still apply.
|
||||
_ = json.Unmarshal([]byte(w.SourceConfig), &cfg)
|
||||
}
|
||||
for _, v := range cfg.Volumes {
|
||||
if v.Target == "" {
|
||||
continue
|
||||
}
|
||||
byTarget[v.Target] = store.WorkloadVolume{Source: v.Source, Target: v.Target, Scope: v.Scope, Name: v.Name}
|
||||
}
|
||||
persisted, err := st.ListWorkloadVolumes(w.ID)
|
||||
if err != nil {
|
||||
return nil, err
|
||||
}
|
||||
for _, p := range persisted {
|
||||
byTarget[p.Target] = store.WorkloadVolume{Source: p.Source, Target: p.Target, Scope: p.Scope, Name: p.Name}
|
||||
}
|
||||
return byTarget, nil
|
||||
}
|
||||
|
||||
func skipReason(scope string) string {
|
||||
switch scope {
|
||||
case string(store.VolumeScopeInstance):
|
||||
|
||||
Reference in New Issue
Block a user