| 1 | //go:build linux |
| 2 | |
| 3 | package host |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "errors" |
| 8 | "net" |
| 9 | "net/http" |
| 10 | "os" |
| 11 | "path/filepath" |
| 12 | "reflect" |
| 13 | "sync/atomic" |
| 14 | "testing" |
| 15 | "time" |
| 16 | |
| 17 | "github.com/carlosframework/platform/internal/hostbox" |
| 18 | "github.com/carlosframework/platform/internal/hostclaim" |
| 19 | "github.com/carlosframework/platform/internal/instancectl" |
| 20 | "github.com/carlosframework/platform/internal/rootstate" |
| 21 | ) |
| 22 | |
| 23 | func servingParkSocket(t *testing.T, handler http.Handler) hostbox.JournalCall { |
| 24 | t.Helper() |
| 25 | return rootJournalClient(servingRootSocket(t, handler)) |
| 26 | } |
| 27 | func servingRootSocket(t *testing.T, handler http.Handler) string { |
| 28 | t.Helper() |
| 29 | dir, err := os.MkdirTemp(os.Getenv("TMPDIR"), "serving-park-") |
| 30 | if err != nil { |
| 31 | t.Fatal(err) |
| 32 | } |
| 33 | t.Cleanup(func() { os.RemoveAll(dir) }) |
| 34 | path := filepath.Join(dir, "root.sock") |
| 35 | listener, err := net.Listen("unix", path) |
| 36 | if err != nil { |
| 37 | t.Fatal(err) |
| 38 | } |
| 39 | server := &instancectl.Server{ClaimStop: handler, PeerOK: func(net.Conn) error { return nil }, Logf: func(string, ...any) {}} |
| 40 | ctx, cancel := context.WithCancel(context.Background()) |
| 41 | done := make(chan error, 1) |
| 42 | go func() { done <- server.Serve(ctx, listener) }() |
| 43 | t.Cleanup(func() { |
| 44 | cancel() |
| 45 | if err := <-done; err != nil { |
| 46 | t.Error(err) |
| 47 | } |
| 48 | }) |
| 49 | return path |
| 50 | } |
| 51 | |
| 52 | func servingTransportRoot(t *testing.T, f *servingRootFixture) ClaimRootHTTP { |
| 53 | t.Helper() |
| 54 | ctx, cancel := context.WithCancel(context.Background()) |
| 55 | p := &servingParkRuntime{authority: f.p, state: servingBorrowedState{f.p.ledger.state}, ctx: ctx, cancel: cancel, idle: make(chan struct{})} |
| 56 | t.Cleanup(func() { |
| 57 | cleanup, cancel := context.WithTimeout(context.Background(), 5*time.Second) |
| 58 | defer cancel() |
| 59 | if err := p.Close(cleanup); err != nil { |
| 60 | t.Error(err) |
| 61 | } |
| 62 | }) |
| 63 | h := f.p.root |
| 64 | h.servingParks = p |
| 65 | h.Stop.Coordinator.RetainNoProcess = p.retainNoProcess |
| 66 | return h |
| 67 | } |
| 68 | |
| 69 | func TestServingParkSocketRetiresAndCompletes(t *testing.T) { |
| 70 | for _, legacy := range []bool{false, true} { |
| 71 | t.Run(map[bool]string{false: "generation", true: "legacy"}[legacy], func(t *testing.T) { |
| 72 | var f *servingRootFixture |
| 73 | if legacy { |
| 74 | f, _ = newServingLegacyRootFixture(t, false) |
| 75 | } else { |
| 76 | f = newServingRootFixture(t, true) |
| 77 | } |
| 78 | h := servingTransportRoot(t, f) |
| 79 | call := servingParkSocket(t, h) |
| 80 | var exact hostbox.JournalRequest |
| 81 | client := hostbox.ServingParkClient{Call: func(ctx context.Context, q hostbox.JournalRequest) (hostbox.JournalResponse, error) { |
| 82 | exact = q |
| 83 | return call(ctx, q) |
| 84 | }} |
| 85 | var a servingParkAuthorization |
| 86 | var err error |
| 87 | if legacy { |
| 88 | a, err = client.ParkLegacy(t.Context(), f.current, servingLegacyPredecessor{}) |
| 89 | } else { |
| 90 | a, err = client.Park(t.Context(), f.current, f.prior) |
| 91 | } |
| 92 | if err != nil { |
| 93 | t.Fatal(err) |
| 94 | } |
| 95 | parkRequest := exact |
| 96 | r, err := f.p.ledger.read(a.ID) |
| 97 | if err != nil || !r.Parked || r.Complete { |
| 98 | t.Fatalf("park: %+v %v", r, err) |
| 99 | } |
| 100 | if _, err = os.Stat(f.db); !errors.Is(err, os.ErrNotExist) { |
| 101 | t.Fatal("source survived successful park", err) |
| 102 | } |
| 103 | f.tags(f.current.Stamp) |
| 104 | if err = client.Complete(t.Context(), a.ID, f.current); err != nil { |
| 105 | t.Fatal(err) |
| 106 | } |
| 107 | r, err = f.p.ledger.read(a.ID) |
| 108 | if err != nil || !r.Complete { |
| 109 | t.Fatalf("completion: %+v %v", r, err) |
| 110 | } |
| 111 | if !legacy { |
| 112 | again, err := client.Park(t.Context(), f.current, f.prior) |
| 113 | if err != nil || again != a { |
| 114 | t.Fatalf("fresh request after completion: %+v %v", again, err) |
| 115 | } |
| 116 | if exact.ID == parkRequest.ID { |
| 117 | t.Fatal("fresh request reused cached transport reply") |
| 118 | } |
| 119 | } |
| 120 | before, err := f.disk.Read("retirement-" + a.ID + ".json") |
| 121 | if err != nil { |
| 122 | t.Fatal(err) |
| 123 | } |
| 124 | // A replay returns the immutable reply even after restore writes the |
| 125 | // successor. It must not inspect or park those new bytes. |
| 126 | if err = os.WriteFile(f.db, []byte("successor"), 0600); err != nil { |
| 127 | t.Fatal(err) |
| 128 | } |
| 129 | out, err := call(t.Context(), parkRequest) |
| 130 | if err != nil || out.Error != "" || out.ServingPark == nil || *out.ServingPark != a { |
| 131 | t.Fatalf("replay: %+v %v", out, err) |
| 132 | } |
| 133 | after, err := f.disk.Read("retirement-" + a.ID + ".json") |
| 134 | if err != nil || string(before) != string(after) { |
| 135 | t.Fatal("replay changed retirement") |
| 136 | } |
| 137 | body, _ := os.ReadFile(f.db) |
| 138 | if string(body) != "successor" { |
| 139 | t.Fatal("replay touched successor") |
| 140 | } |
| 141 | }) |
| 142 | } |
| 143 | } |
| 144 | |
| 145 | func TestServingParkSocketLostReplyReplaysExactRequest(t *testing.T) { |
| 146 | f := newServingRootFixture(t, true) |
| 147 | h := servingTransportRoot(t, f) |
| 148 | var requests atomic.Int32 |
| 149 | call := servingParkSocket(t, http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { |
| 150 | if requests.Add(1) == 1 { |
| 151 | // Execute against the real connection context, then drop the reply. |
| 152 | h.ServeHTTP(discardServingReply{HeaderMap: make(http.Header)}, r) |
| 153 | conn, _, err := w.(http.Hijacker).Hijack() |
| 154 | if err == nil { |
| 155 | conn.Close() |
| 156 | } |
| 157 | return |
| 158 | } |
| 159 | h.ServeHTTP(w, r) |
| 160 | })) |
| 161 | client := hostbox.ServingParkClient{Call: call} |
| 162 | a, err := client.Park(t.Context(), f.current, f.prior) |
| 163 | if err != nil || requests.Load() != 2 { |
| 164 | t.Fatalf("lost reply: %+v %v requests=%d", a, err, requests.Load()) |
| 165 | } |
| 166 | r, err := f.p.ledger.read(a.ID) |
| 167 | if err != nil || !r.Parked { |
| 168 | t.Fatalf("park: %+v %v", r, err) |
| 169 | } |
| 170 | } |
| 171 | |
| 172 | type discardServingReply struct{ HeaderMap http.Header } |
| 173 | |
| 174 | func (w discardServingReply) Header() http.Header { return w.HeaderMap } |
| 175 | func (discardServingReply) WriteHeader(int) {} |
| 176 | func (discardServingReply) Write(b []byte) (int, error) { return len(b), nil } |
| 177 | |
| 178 | func TestServingParkSocketAuthorityGates(t *testing.T) { |
| 179 | for _, gate := range []string{"peer", "census", "enrollment", "revision", "box", "quarantine"} { |
| 180 | t.Run(gate, func(t *testing.T) { |
| 181 | f := newServingRootFixture(t, true) |
| 182 | h := servingTransportRoot(t, f) |
| 183 | var refuse atomic.Bool |
| 184 | auth := h.Stop.Authenticate |
| 185 | h.Stop.Authenticate = func(ctx context.Context, c net.Conn) (StopPeer, error) { |
| 186 | if refuse.Load() && gate == "peer" { |
| 187 | return StopPeer{}, ErrClaimStopIdentity |
| 188 | } |
| 189 | return auth(ctx, c) |
| 190 | } |
| 191 | h.RequireNewWork = func(context.Context) error { |
| 192 | if refuse.Load() && gate == "census" { |
| 193 | return hostbox.ErrChanged |
| 194 | } |
| 195 | return nil |
| 196 | } |
| 197 | h.InventoryActor = func(context.Context) (hostbox.Actor, error) { |
| 198 | a := f.actor |
| 199 | if refuse.Load() { |
| 200 | switch gate { |
| 201 | case "enrollment": |
| 202 | a.Enrollment = "other" |
| 203 | case "revision": |
| 204 | a.Revision = 0 |
| 205 | case "box": |
| 206 | a.Box = "other" |
| 207 | case "quarantine": |
| 208 | a.Quarantined = true |
| 209 | } |
| 210 | } |
| 211 | return a, nil |
| 212 | } |
| 213 | call := servingParkSocket(t, h) |
| 214 | var q hostbox.JournalRequest |
| 215 | client := hostbox.ServingParkClient{Call: func(ctx context.Context, in hostbox.JournalRequest) (hostbox.JournalResponse, error) { |
| 216 | q = in |
| 217 | return call(ctx, in) |
| 218 | }} |
| 219 | a, err := client.Park(t.Context(), f.current, f.prior) |
| 220 | if err != nil { |
| 221 | t.Fatal(err) |
| 222 | } |
| 223 | before, _ := f.disk.Read("retirement-" + a.ID + ".json") |
| 224 | refuse.Store(true) |
| 225 | out, err := call(t.Context(), q) |
| 226 | if err == nil && out.Error == "" { |
| 227 | t.Fatal("replay bypassed authority", gate) |
| 228 | } |
| 229 | after, _ := f.disk.Read("retirement-" + a.ID + ".json") |
| 230 | if !reflect.DeepEqual(before, after) { |
| 231 | t.Fatal("refusal mutated record") |
| 232 | } |
| 233 | }) |
| 234 | } |
| 235 | } |
| 236 | |
| 237 | type servingCloseStore struct { |
| 238 | rootstate.Store |
| 239 | closed atomic.Bool |
| 240 | } |
| 241 | |
| 242 | func (s *servingCloseStore) Close() error { s.closed.Store(true); return nil } |
| 243 | |
| 244 | func TestServingParkRuntimeJoinBeforePrivateStateClose(t *testing.T) { |
| 245 | s := &servingCloseStore{} |
| 246 | ctx, cancel := context.WithCancel(context.Background()) |
| 247 | p := &servingParkRuntime{state: s, ctx: ctx, cancel: cancel, idle: make(chan struct{})} |
| 248 | call, done, err := p.begin(t.Context()) |
| 249 | if err != nil { |
| 250 | t.Fatal(err) |
| 251 | } |
| 252 | deadline, stop := context.WithCancel(context.Background()) |
| 253 | stop() |
| 254 | if err = p.Close(deadline); !errors.Is(err, context.Canceled) || s.closed.Load() { |
| 255 | t.Fatal("closed state before join", err) |
| 256 | } |
| 257 | select { |
| 258 | case <-call.Done(): |
| 259 | case <-time.After(time.Second): |
| 260 | t.Fatal("active call not canceled") |
| 261 | } |
| 262 | if _, _, err = p.begin(t.Context()); err == nil { |
| 263 | t.Fatal("new call admitted during close") |
| 264 | } |
| 265 | done() |
| 266 | if err = p.Close(t.Context()); err != nil || !s.closed.Load() { |
| 267 | t.Fatal("join did not close store", err) |
| 268 | } |
| 269 | if err = p.Close(t.Context()); err != nil { |
| 270 | t.Fatal(err) |
| 271 | } |
| 272 | } |
| 273 | |
| 274 | func TestServingParkRuntimeRetainsNoProcess(t *testing.T) { |
| 275 | // The production lifecycle wrapper must retain the same root observation |
| 276 | // as the direct coordinator fixture, and refuse after shutdown. |
| 277 | f, _ := newServingLegacyRootFixture(t, false) |
| 278 | h := servingTransportRoot(t, f) |
| 279 | saved, err := f.p.ledger.legacyStop(f.current) |
| 280 | if err != nil { |
| 281 | t.Fatal(err) |
| 282 | } |
| 283 | // Retention rechecks the original claim, so it refuses after admission. |
| 284 | if err = h.Stop.Coordinator.RetainNoProcess(t.Context(), saved.Observed, saved.Result); err == nil { |
| 285 | t.Fatal("retained stale preadmission observation") |
| 286 | } |
| 287 | if err = h.servingParks.Close(t.Context()); err != nil { |
| 288 | t.Fatal(err) |
| 289 | } |
| 290 | if err = h.Stop.Coordinator.RetainNoProcess(t.Context(), hostclaim.Observed{}, StopResult{}); err == nil { |
| 291 | t.Fatal("closed runtime retained evidence") |
| 292 | } |
| 293 | } |
| 294 | |
| 295 | func TestServingParkProductionConstructorRequiresPrivateDependencies(t *testing.T) { |
| 296 | f := newServingRootFixture(t, true) |
| 297 | if _, err := newServingParkRuntime(f.p.root, nil, nil); err == nil { |
| 298 | t.Fatal("missing private dependencies accepted") |
| 299 | } |
| 300 | // Production FS construction fixes root ownership and the nosymfollow |
| 301 | // check. This test does not claim to open a privileged rootstate store. |
| 302 | p, err := newServingParkRuntime(f.p.root, f.disk, plan9D5Fixture(t).owners) |
| 303 | if err != nil { |
| 304 | t.Fatal(err) |
| 305 | } |
| 306 | defer p.Close(t.Context()) |
| 307 | fs := p.authority.(*servingParkRoot).ledger.fs.(*servingParkFS) |
| 308 | if fs.owner != 0 || fs.mount == nil || p.authority.(*servingParkRoot).ledger.journal != f.p.root.Journals.Launches { |
| 309 | t.Fatal("production constructor replaced root authority") |
| 310 | } |
| 311 | ctx, cancel := context.WithTimeout(t.Context(), time.Second) |
| 312 | defer cancel() |
| 313 | if err = p.Close(ctx); err != nil { |
| 314 | t.Fatal(err) |
| 315 | } |
| 316 | } |
| 317 | |
| 318 | func TestServingParkProductionBindingRetainsSessionZero(t *testing.T) { |
| 319 | old := openClaimServingParkAuthority |
| 320 | t.Cleanup(func() { openClaimServingParkAuthority = old }) |
| 321 | var closed *servingCloseStore |
| 322 | f, _ := newServingLegacyRootFixture(t, false, func(f *servingRootFixture, coordinator *ClaimStopCoordinator, request StopRequest) { |
| 323 | closed = &servingCloseStore{Store: f.disk} |
| 324 | openClaimServingParkAuthority = func() (rootstate.Store, error) { return closed, nil } |
| 325 | h, err := withServingParkRoot(f.p.root, f.p.empty) |
| 326 | if err != nil { |
| 327 | t.Fatal(err) |
| 328 | } |
| 329 | t.Cleanup(func() { |
| 330 | if err := h.servingParks.Close(context.Background()); err != nil { |
| 331 | t.Error(err) |
| 332 | } |
| 333 | }) |
| 334 | // This repeats the real preadmission coordinator path through production |
| 335 | // binding. The original private no-process record must remain identical. |
| 336 | key := servingLegacyKey("legacy-stop", servingArtifactIdentity{Host: request.Host, Box: request.Box, Incarnation: request.Incarnation}) |
| 337 | before, err := f.disk.Read(key) |
| 338 | if err != nil { |
| 339 | t.Fatal(err) |
| 340 | } |
| 341 | coordinator.RetainNoProcess = h.Stop.Coordinator.RetainNoProcess |
| 342 | if _, err = coordinator.FenceAndStop(t.Context(), request); err != nil { |
| 343 | t.Fatal(err) |
| 344 | } |
| 345 | after, err := f.disk.Read(key) |
| 346 | if err != nil || !reflect.DeepEqual(before, after) { |
| 347 | t.Fatal("production retention changed original evidence", err) |
| 348 | } |
| 349 | if closed.closed.Load() { |
| 350 | t.Fatal("private store closed during service") |
| 351 | } |
| 352 | if err = h.servingParks.Close(t.Context()); err != nil { |
| 353 | t.Fatal(err) |
| 354 | } |
| 355 | }) |
| 356 | if f == nil || closed == nil || !closed.closed.Load() { |
| 357 | t.Fatal("production private store leaked") |
| 358 | } |
| 359 | bad := &servingCloseStore{Store: f.disk} |
| 360 | openClaimServingParkAuthority = func() (rootstate.Store, error) { return bad, nil } |
| 361 | if _, err := withServingParkRoot(ClaimRootHTTP{}, nil); err == nil || !bad.closed.Load() { |
| 362 | t.Fatal("failed binding leaked private store", err) |
| 363 | } |
| 364 | } |
| 365 | |
| 366 | type servingBorrowedState struct{ rootstate.Store } |
| 367 | |
| 368 | func (servingBorrowedState) Close() error { return nil } |
| 369 | |
amadan