| 1 | //go:build linux && rootrunner |
| 2 | |
| 3 | package host |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "crypto/ed25519" |
| 8 | "crypto/sha256" |
| 9 | "encoding/hex" |
| 10 | "encoding/json" |
| 11 | "fmt" |
| 12 | "net" |
| 13 | "net/http" |
| 14 | "net/url" |
| 15 | "os" |
| 16 | "os/exec" |
| 17 | "path/filepath" |
| 18 | "strings" |
| 19 | "testing" |
| 20 | "time" |
| 21 | |
| 22 | "github.com/carlosframework/platform/internal/hostbox" |
| 23 | "github.com/carlosframework/platform/internal/hostclaim" |
| 24 | execlease "github.com/carlosframework/platform/internal/hostexec" |
| 25 | "github.com/carlosframework/platform/internal/hostguard" |
| 26 | "github.com/carlosframework/platform/internal/instancectl" |
| 27 | "github.com/carlosframework/platform/internal/provision" |
| 28 | "github.com/carlosframework/platform/internal/registry" |
| 29 | "github.com/carlosframework/platform/internal/release" |
| 30 | "github.com/carlosframework/platform/internal/rollout" |
| 31 | "github.com/carlosframework/platform/internal/rootstate" |
| 32 | ) |
| 33 | |
| 34 | func plan14InstalledTenant() { |
| 35 | if os.Getenv("PLAN14_TENANT_FAIL") == "1" { |
| 36 | os.Exit(1) |
| 37 | } |
| 38 | socket := os.Getenv("PLAN14_TENANT_SOCKET") |
| 39 | if socket == "" { |
| 40 | os.Exit(2) |
| 41 | } |
| 42 | network := "unix" |
| 43 | if addr := os.Getenv("PLAN14_TENANT_ADDR"); addr != "" { |
| 44 | network, socket = "tcp", addr |
| 45 | } |
| 46 | if network == "unix" { |
| 47 | // The payload owns its socket, like an ordinary Unix HTTP server. A |
| 48 | // process killed at exit may leave that socket pathname behind. |
| 49 | if st, err := os.Lstat(socket); err == nil { |
| 50 | if st.Mode()&os.ModeSocket == 0 || os.Remove(socket) != nil { |
| 51 | os.Exit(1) |
| 52 | } |
| 53 | } else if !os.IsNotExist(err) { |
| 54 | os.Exit(1) |
| 55 | } |
| 56 | } |
| 57 | ln, err := net.Listen(network, socket) |
| 58 | if err != nil { |
| 59 | fmt.Fprintln(os.Stderr, err) |
| 60 | os.Exit(1) |
| 61 | } |
| 62 | if err := http.Serve(ln, http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { |
| 63 | if target := r.URL.Query().Get("dial"); target != "" { |
| 64 | conn, err := net.DialTimeout("tcp", target, 200*time.Millisecond) |
| 65 | if err != nil { |
| 66 | w.WriteHeader(http.StatusForbidden) |
| 67 | return |
| 68 | } |
| 69 | conn.Close() |
| 70 | w.WriteHeader(http.StatusNoContent) |
| 71 | return |
| 72 | } |
| 73 | w.Write([]byte("running")) |
| 74 | })); err != nil { |
| 75 | os.Exit(1) |
| 76 | } |
| 77 | } |
| 78 | |
| 79 | func TestPlan14InstalledClaimLaunchRootRunner(t *testing.T) { |
| 80 | plan14InstalledClaimLaunch(t, "") |
| 81 | } |
| 82 | |
| 83 | func TestPlan14NamedClaimLaunchRootRunner(t *testing.T) { |
| 84 | for _, unit := range []string{"plan14-named.service", "plan14-socket.service", "backing"} { |
| 85 | t.Run(unit, func(t *testing.T) { plan14InstalledClaimLaunch(t, unit) }) |
| 86 | } |
| 87 | } |
| 88 | |
| 89 | func TestPlan14RootRuntimeInstalledCompletion(t *testing.T) { plan14InstalledClaimLaunch(t, "", true) } |
| 90 | |
| 91 | func TestPlan14RootRuntimeInstalledFailure(t *testing.T) { plan14InstalledClaimLaunch(t, "", false) } |
| 92 | |
| 93 | // This witness permits loopback HTTP between the runner and its named TCP |
| 94 | // payloads. The payloads make no remote network requests. |
| 95 | func TestPlan14BoundRouteProgramRootRunner(t *testing.T) { |
| 96 | for _, unit := range []string{"tcp", "socket-program", "backing-program"} { |
| 97 | t.Run(unit, func(t *testing.T) { plan14InstalledClaimLaunch(t, unit) }) |
| 98 | } |
| 99 | } |
| 100 | |
| 101 | func TestPlan14MaterializedBoundRouteRootRunner(t *testing.T) { |
| 102 | for _, unit := range []string{"tcp", "socket-program", "backing-program"} { |
| 103 | t.Run(unit, func(t *testing.T) { plan14InstalledClaimLaunch(t, "materialized:"+unit) }) |
| 104 | } |
| 105 | } |
| 106 | |
| 107 | func TestPlan14MigratedBoundRouteRootRunner(t *testing.T) { |
| 108 | for _, unit := range []string{"tcp", "socket-program", "backing-program"} { |
| 109 | t.Run(unit, func(t *testing.T) { plan14InstalledClaimLaunch(t, "migrated:materialized:"+unit) }) |
| 110 | } |
| 111 | } |
| 112 | |
| 113 | func TestPlan14BoundNetworkEdgeRootRunner(t *testing.T) { |
| 114 | for _, unit := range []string{"tcp", "socket-program", "backing-program"} { |
| 115 | t.Run(unit, func(t *testing.T) { plan14InstalledClaimLaunch(t, "isolation:materialized:"+unit) }) |
| 116 | } |
| 117 | } |
| 118 | |
| 119 | func TestPlan14ConsoleLibraryRootRunner(t *testing.T) { |
| 120 | plan14InstalledClaimLaunch(t, "console:materialized:tcp") |
| 121 | } |
| 122 | |
| 123 | var plan14ConsoleExecutable, plan14WoodstarExecutable []byte |
| 124 | |
| 125 | func TestPlan14RealConsoleRootRunnerPayload(t *testing.T) { |
| 126 | if len(plan14ConsoleExecutable) == 0 { |
| 127 | t.Fatal("build the real executable with scripts/plan14-console-root-runner.sh") |
| 128 | } |
| 129 | plan14InstalledClaimLaunch(t, "binary:console:materialized:tcp") |
| 130 | } |
| 131 | |
| 132 | func TestPlan14RealWoodstarRootRunnerPayload(t *testing.T) { |
| 133 | if len(plan14WoodstarExecutable) == 0 { |
| 134 | t.Fatal("set PLAN14_WOODSTAR_SOURCE when running scripts/plan14-console-root-runner.sh") |
| 135 | } |
| 136 | plan14InstalledClaimLaunch(t, "woodstar:materialized:tcp") |
| 137 | } |
| 138 | |
| 139 | func TestPlan14NamedStartupDefersReplacementRootRunner(t *testing.T) { |
| 140 | for _, mode := range []string{"tcp", "socket-program", "backing-program"} { |
| 141 | t.Run(mode, func(t *testing.T) { plan14InstalledClaimLaunch(t, "defer:migrated:materialized:"+mode) }) |
| 142 | } |
| 143 | } |
| 144 | |
| 145 | func TestPlan14OrdinaryUnitAuthorizationRootRunner(t *testing.T) { |
| 146 | plan14InstalledClaimLaunch(t, "ordinary-verified") |
| 147 | } |
| 148 | |
| 149 | func plan14InstalledClaimLaunch(t *testing.T, boundUnit string, runtimeUpdate ...bool) { |
| 150 | verifiedOrdinary := boundUnit == "ordinary-verified" |
| 151 | if verifiedOrdinary { |
| 152 | boundUnit = "" |
| 153 | } |
| 154 | woodstar := strings.HasPrefix(boundUnit, "woodstar:") |
| 155 | boundUnit = strings.TrimPrefix(boundUnit, "woodstar:") |
| 156 | deferReplacement := strings.HasPrefix(boundUnit, "defer:") |
| 157 | boundUnit = strings.TrimPrefix(boundUnit, "defer:") |
| 158 | consoleBinary := strings.HasPrefix(boundUnit, "binary:") |
| 159 | boundUnit = strings.TrimPrefix(boundUnit, "binary:") |
| 160 | realConsole := strings.HasPrefix(boundUnit, "console:") |
| 161 | boundUnit = strings.TrimPrefix(boundUnit, "console:") |
| 162 | isolation := strings.HasPrefix(boundUnit, "isolation:") |
| 163 | boundUnit = strings.TrimPrefix(boundUnit, "isolation:") |
| 164 | startup := strings.HasPrefix(boundUnit, "startup:") |
| 165 | boundUnit = strings.TrimPrefix(boundUnit, "startup:") |
| 166 | combined := strings.HasPrefix(boundUnit, "combined:") |
| 167 | boundUnit = strings.TrimPrefix(boundUnit, "combined:") |
| 168 | migrated := strings.HasPrefix(boundUnit, "migrated:") |
| 169 | boundUnit = strings.TrimPrefix(boundUnit, "migrated:") |
| 170 | materialized := strings.HasPrefix(boundUnit, "materialized:") |
| 171 | boundUnit = strings.TrimPrefix(boundUnit, "materialized:") |
| 172 | programMode := boundUnit == "tcp" || boundUnit == "socket-program" || boundUnit == "backing-program" |
| 173 | tcpMode := boundUnit == "tcp" || boundUnit == "backing-program" |
| 174 | if boundUnit == "tcp" { |
| 175 | boundUnit = "plan14-tcp.service" |
| 176 | } |
| 177 | if boundUnit == "socket-program" { |
| 178 | boundUnit = "plan14-socket.service" |
| 179 | } |
| 180 | if boundUnit == "backing-program" { |
| 181 | boundUnit = "backing" |
| 182 | } |
| 183 | if dir := os.Getenv("PLAN14_INSTALLED_RETIRE_CHILD"); dir != "" { |
| 184 | plan14InstalledRetire(t, dir) |
| 185 | return |
| 186 | } |
| 187 | dir := generationRootRunnerDir(t) |
| 188 | t.Setenv("TMPDIR", dir) |
| 189 | uid, gid := generationRunnerRegistryUser(t) |
| 190 | ctl := func(ctx context.Context, args ...string) (string, error) { |
| 191 | out, err := exec.CommandContext(ctx, "systemctl", args...).CombinedOutput() |
| 192 | return strings.TrimSpace(string(out)), err |
| 193 | } |
| 194 | // These fixture paths must be absent. Install only this runner's units and |
| 195 | // filesystem roots; retain the deployed unit's hardening and launch entrypoint. |
| 196 | for _, p := range []string{"/data", "/opt/carlos", "/run/carlos", "/run/carlos-control", "/etc/carlos"} { |
| 197 | if err := os.Mkdir(p, 0755); err != nil { |
| 198 | t.Fatal("fixture root already exists", p, err) |
| 199 | } |
| 200 | p := p |
| 201 | t.Cleanup(func() { |
| 202 | if err := os.RemoveAll(p); err != nil { |
| 203 | t.Error(err) |
| 204 | } |
| 205 | }) |
| 206 | } |
| 207 | binary, err := os.Executable() |
| 208 | if err != nil { |
| 209 | t.Fatal(err) |
| 210 | } |
| 211 | body, err := os.ReadFile(binary) |
| 212 | if err != nil { |
| 213 | t.Fatal(err) |
| 214 | } |
| 215 | if err := os.WriteFile(instancectl.SystemBinary, body, 0755); err != nil { |
| 216 | t.Fatal(err) |
| 217 | } |
| 218 | f, _, rec := generationLaunchFixture(t) |
| 219 | t.Setenv("CARLOS_DATA_DIR", "/data") |
| 220 | t.Setenv("CARLOS_OPT_DIR", "/opt/carlos") |
| 221 | host := f.target.Host |
| 222 | var endpoint string |
| 223 | if tcpMode { |
| 224 | ln, err := net.Listen("tcp", "127.0.0.1:0") |
| 225 | if err != nil { |
| 226 | t.Fatal(err) |
| 227 | } |
| 228 | endpoint = ln.Addr().String() |
| 229 | ln.Close() |
| 230 | } |
| 231 | socketEndpoint := filepath.Join(provision.InstanceRunDir(host), "app.sock") |
| 232 | if materialized && !tcpMode { |
| 233 | socketEndpoint = "/run/carlos/woodstar-test.sock" |
| 234 | } |
| 235 | program := claimUnitProgram{Schema: 1, Host: host, Args: []string{"plan14-tenant"}, Env: []string{"PLAN14_TENANT_SOCKET=" + socketEndpoint}} |
| 236 | if tcpMode { |
| 237 | program.Env = append(program.Env, "PLAN14_TENANT_ADDR="+endpoint) |
| 238 | } |
| 239 | var checkService func(*http.Client) |
| 240 | if realConsole { |
| 241 | program, checkService = plan14ConsoleProgram(t, host, endpoint) |
| 242 | if consoleBinary { |
| 243 | if len(plan14ConsoleExecutable) == 0 { |
| 244 | t.Fatal("real console executable missing") |
| 245 | } |
| 246 | program = plan14ConsoleBinaryProgram(t, program) |
| 247 | } |
| 248 | } |
| 249 | if woodstar { |
| 250 | if len(plan14WoodstarExecutable) == 0 { |
| 251 | t.Fatal("real Woodstar executable missing") |
| 252 | } |
| 253 | program, checkService = plan14WoodstarProgram(t, host, endpoint) |
| 254 | } |
| 255 | healthURL := "http://tenant/" |
| 256 | if woodstar { |
| 257 | healthURL += "healthz" |
| 258 | } |
| 259 | programRaw, _ := json.Marshal(program) |
| 260 | programSum := sha256.Sum256(programRaw) |
| 261 | if boundUnit == "backing" { |
| 262 | boundUnit = "carlos-app@" + host + ".service" |
| 263 | } |
| 264 | for _, p := range []string{provision.InstanceDir("/data", host), provision.InstanceRunDir(host), "/run/carlos-control/exe", "/run/carlos/logs"} { |
| 265 | if err := os.MkdirAll(p, 0755); err != nil { |
| 266 | t.Fatal(err) |
| 267 | } |
| 268 | } |
| 269 | for _, p := range []string{provision.InstanceDir("/data", host), provision.InstanceRunDir(host)} { |
| 270 | if err := os.Chown(p, uid, gid); err != nil { |
| 271 | t.Fatal(err) |
| 272 | } |
| 273 | } |
| 274 | if materialized || verifiedOrdinary { |
| 275 | plan14MaterializedDependencies(t, ctl) |
| 276 | } |
| 277 | if verifiedOrdinary { |
| 278 | path := "/etc/systemd/system/carlos-edge.service" |
| 279 | if _, err := os.Lstat(path); !os.IsNotExist(err) { |
| 280 | t.Fatal("existing edge", err) |
| 281 | } |
| 282 | if err := os.WriteFile(path, []byte("[Service]\nType=oneshot\nExecStart=/usr/bin/true\nRemainAfterExit=yes\n"), 0644); err != nil { |
| 283 | t.Fatal(err) |
| 284 | } |
| 285 | t.Cleanup(func() { |
| 286 | ctl(context.Background(), "stop", "carlos-edge.service") |
| 287 | os.Remove(path) |
| 288 | ctl(context.Background(), "daemon-reload") |
| 289 | }) |
| 290 | } |
| 291 | template, err := unitsFS.ReadFile("units/carlos-instance@.service") |
| 292 | if err != nil { |
| 293 | t.Fatal(err) |
| 294 | } |
| 295 | var lines []string |
| 296 | for _, line := range strings.Split(string(template), "\n") { |
| 297 | key, _, _ := strings.Cut(line, "=") |
| 298 | switch key { |
| 299 | case "BindsTo", "Requires", "After": |
| 300 | continue |
| 301 | case "StandardOutput": |
| 302 | line = "StandardOutput=journal" |
| 303 | case "ExecStart": |
| 304 | if programMode { |
| 305 | line = "ExecStart=/usr/bin/false" |
| 306 | } |
| 307 | } |
| 308 | lines = append(lines, line) |
| 309 | } |
| 310 | if tcpMode { |
| 311 | lines = append(lines, "[Service]", "IPAddressAllow=localhost") |
| 312 | } |
| 313 | unit := instancectl.UnitName(instancectl.RoleRun, host) |
| 314 | unitPath := "/run/systemd/system/carlos-instance@.service" |
| 315 | if verifiedOrdinary { |
| 316 | unitPath = "/etc/systemd/system/carlos-instance@.service" |
| 317 | lines = strings.Split(string(template), "\n") |
| 318 | } |
| 319 | if boundUnit != "" { |
| 320 | unit = boundUnit |
| 321 | unitPath = filepath.Join("/run/systemd/system", unit) |
| 322 | for i := range lines { |
| 323 | lines[i] = strings.ReplaceAll(strings.ReplaceAll(lines[i], "%i", provision.InstanceID(host)), "%I", provision.InstanceID(host)) |
| 324 | } |
| 325 | f.req.Reservation.Group.RelativePath = instancectl.BoundClaimUnitControlGroup(host, unit) |
| 326 | } |
| 327 | if materialized { |
| 328 | unitPath = filepath.Join("/etc/systemd/system", unit) |
| 329 | } |
| 330 | legacyDefinition := []byte("[Service]\nType=simple\nUser=carlos\nKillMode=control-group\nExecStart=/bin/sh -c 'sleep infinity & wait'\n") |
| 331 | if migrated && !tcpMode { |
| 332 | if err := os.Chown("/run/carlos", uid, gid); err != nil { |
| 333 | t.Fatal(err) |
| 334 | } |
| 335 | legacyDefinition = []byte("[Service]\nType=simple\nUser=carlos\nKillMode=control-group\nEnvironment=PLAN14_TENANT_SOCKET=" + socketEndpoint + "\nExecStart=/opt/carlos/carlos plan14-tenant\n") |
| 336 | } |
| 337 | if migrated { |
| 338 | if err := os.WriteFile(unitPath, legacyDefinition, 0644); err != nil { |
| 339 | t.Fatal(err) |
| 340 | } |
| 341 | } |
| 342 | if !materialized { |
| 343 | out, err := os.OpenFile(unitPath, os.O_CREATE|os.O_EXCL|os.O_WRONLY, 0644) |
| 344 | if err != nil { |
| 345 | t.Fatal(err) |
| 346 | } |
| 347 | if _, err := out.WriteString(strings.Join(lines, "\n")); err != nil { |
| 348 | t.Fatal(err) |
| 349 | } |
| 350 | out.Close() |
| 351 | } |
| 352 | t.Cleanup(func() { |
| 353 | ctl(context.Background(), "stop", unit) |
| 354 | ctl(context.Background(), "stop", instancectl.SliceName(host)) |
| 355 | ctl(context.Background(), "reset-failed", unit) |
| 356 | os.Remove(unitPath) |
| 357 | os.RemoveAll(filepath.Join("/run/systemd/system", unit+".d")) |
| 358 | os.RemoveAll(filepath.Join("/run/systemd/system", instancectl.SliceName(host)+".d")) |
| 359 | ctl(context.Background(), "daemon-reload") |
| 360 | }) |
| 361 | if out, err := ctl(t.Context(), "daemon-reload"); err != nil { |
| 362 | t.Fatal(out, err) |
| 363 | } |
| 364 | if migrated { |
| 365 | if out, err := ctl(t.Context(), "start", unit); err != nil { |
| 366 | t.Fatal(out, err) |
| 367 | } |
| 368 | if !tcpMode { |
| 369 | deadline := time.Now().Add(5 * time.Second) |
| 370 | for { |
| 371 | conn, err := net.DialTimeout("unix", socketEndpoint, time.Second) |
| 372 | if err == nil { |
| 373 | conn.Close() |
| 374 | break |
| 375 | } |
| 376 | if time.Now().After(deadline) { |
| 377 | t.Fatal("legacy socket did not serve", err) |
| 378 | } |
| 379 | time.Sleep(20 * time.Millisecond) |
| 380 | } |
| 381 | } |
| 382 | } |
| 383 | // Use a real controller lifetime, which must be joined before exact release. |
| 384 | controller := exec.Command("/bin/sleep", "240") |
| 385 | if err := controller.Start(); err != nil { |
| 386 | t.Fatal(err) |
| 387 | } |
| 388 | joined := false |
| 389 | join := func() { |
| 390 | if !joined { |
| 391 | controller.Process.Kill() |
| 392 | controller.Wait() |
| 393 | joined = true |
| 394 | } |
| 395 | } |
| 396 | t.Cleanup(join) |
| 397 | f.peer.PID = controller.Process.Pid |
| 398 | f.peer.ProcessStart, err = claimLaunchProcessStart("/proc", f.peer.PID) |
| 399 | if err != nil { |
| 400 | t.Fatal(err) |
| 401 | } |
| 402 | launch, err := f.journal.Read(t.Context(), f.target) |
| 403 | if err != nil { |
| 404 | t.Fatal(err) |
| 405 | } |
| 406 | boot, err := os.ReadFile("/proc/sys/kernel/random/boot_id") |
| 407 | if err != nil { |
| 408 | t.Fatal(err) |
| 409 | } |
| 410 | f.target.BootID, f.peer.BootID = strings.TrimSpace(string(boot)), strings.TrimSpace(string(boot)) |
| 411 | f.req.Reservation.Target = f.target |
| 412 | launch.Target = f.target |
| 413 | f.claims.set(func(c *hostclaim.Claim) { c.Holder.WakeRetry.Attempt.BootID = f.peer.BootID }) |
| 414 | launch.ControllerPID, launch.ProcessStart = f.peer.PID, f.peer.ProcessStart |
| 415 | // Open a fresh journal with the controller identity production will check. |
| 416 | journal, err := hostbox.OpenLaunchJournal(filepath.Join(dir, "launch.db")) |
| 417 | if err != nil { |
| 418 | t.Fatal(err) |
| 419 | } |
| 420 | t.Cleanup(func() { journal.Close() }) |
| 421 | if err := journal.Prelaunch(t.Context(), launch); err != nil { |
| 422 | t.Fatal(err) |
| 423 | } |
| 424 | f.journal, f.root.Journal = journal, journal |
| 425 | state, err := rootstate.Open(filepath.Join(dir, "fences"), true, false) |
| 426 | if err != nil { |
| 427 | t.Fatal(err) |
| 428 | } |
| 429 | defer state.Close() |
| 430 | old, err := f.root.Guard.Root.Installation(t.Context(), host, hostguard.Captured{Incarnation: f.target.Incarnation, Stamp: f.target.Stamp}) |
| 431 | if err != nil { |
| 432 | t.Fatal(err) |
| 433 | } |
| 434 | f.root.Guard.Root = hostguard.NewRootAuthority(state) |
| 435 | if err := f.root.Guard.EstablishRoot(t.Context(), old, func(*hostclaim.Claim) error { return nil }); err != nil { |
| 436 | t.Fatal(err) |
| 437 | } |
| 438 | if boundUnit != "" { |
| 439 | bindings, err := rootstate.Open(filepath.Join(dir, "unit-bindings"), true, false) |
| 440 | if err != nil { |
| 441 | t.Fatal(err) |
| 442 | } |
| 443 | defer bindings.Close() |
| 444 | shape := claimUnitShape{Host: host, Account: old.Account, App: old.App, Kind: "instance", Unit: boundUnit, Socket: provision.NestedSocket(host), DBPath: provision.NestedDBPath("/data", host)} |
| 445 | if strings.HasPrefix(boundUnit, "carlos-app@") { |
| 446 | shape.Backing = "unit" |
| 447 | } |
| 448 | if migrated { |
| 449 | shape.PredecessorDigest = claimUnitDigest(legacyDefinition) |
| 450 | } |
| 451 | if programMode { |
| 452 | shape.Program = hex.EncodeToString(programSum[:]) |
| 453 | if materialized && !tcpMode { |
| 454 | shape.Socket = socketEndpoint |
| 455 | } |
| 456 | if tcpMode { |
| 457 | shape.Addr, shape.Socket = endpoint, "" |
| 458 | } |
| 459 | if err := os.MkdirAll("/etc/carlos/claim-units", 0700); err != nil { |
| 460 | t.Fatal(err) |
| 461 | } |
| 462 | if err := os.WriteFile(filepath.Join("/etc/carlos/claim-units", host+".json"), programRaw, 0600); err != nil { |
| 463 | t.Fatal(err) |
| 464 | } |
| 465 | programPath := filepath.Join("/etc/carlos/claim-units", host+".json") |
| 466 | if err := os.Chmod(programPath, 0666); err != nil { |
| 467 | t.Fatal(err) |
| 468 | } |
| 469 | if _, err := readClaimUnitProgram(shape); err == nil { |
| 470 | t.Fatal("service-writable program accepted") |
| 471 | } |
| 472 | if err := os.Chmod(programPath, 0600); err != nil { |
| 473 | t.Fatal(err) |
| 474 | } |
| 475 | f.root.unitProgram = readClaimUnitProgram |
| 476 | } |
| 477 | f.root.unitBindings = &claimUnitBindings{state: bindings, installation: func(context.Context, string) (hostguard.RootInstallation, error) { return old, nil }, canonical: func(context.Context, string) (claimUnitShape, error) { return shape, nil }} |
| 478 | if _, err := f.root.unitBindings.establish(t.Context(), host); err != nil { |
| 479 | t.Fatal(err) |
| 480 | } |
| 481 | } |
| 482 | f.root.Clock = execlease.SystemClock{} |
| 483 | f.root.DataRoot = "/data" |
| 484 | f.root.ReadyBudget = 10 * time.Second |
| 485 | f.root.ReadyPoll = 20 * time.Millisecond |
| 486 | f.root.Units = instancectl.ClaimUnits{BoundUnit: boundUnit, BoundHost: func() string { |
| 487 | if boundUnit != "" { |
| 488 | return host |
| 489 | } |
| 490 | return "" |
| 491 | }(), Systemctl: ctl, CgroupRoot: "/sys/fs/cgroup", ExeRoot: "/run/carlos-control/exe", UnitRoot: "/run/systemd/system", DataRoot: "/data", WaitBudget: 20 * time.Second} |
| 492 | if verifiedOrdinary { |
| 493 | oldOpen, oldMaterial, oldShapes := openClaimUnitBindings, openClaimUnitMaterializations, productionClaimUnitShapes |
| 494 | defer func() { |
| 495 | openClaimUnitBindings, openClaimUnitMaterializations, productionClaimUnitShapes = oldOpen, oldMaterial, oldShapes |
| 496 | }() |
| 497 | openClaimUnitBindings = func() (rootstate.Store, error) { |
| 498 | return rootstate.Open(filepath.Join(dir, "ordinary-bindings"), true, false) |
| 499 | } |
| 500 | openClaimUnitMaterializations = func() (rootstate.Store, error) { |
| 501 | return rootstate.Open(filepath.Join(dir, "ordinary-materials"), true, false) |
| 502 | } |
| 503 | productionClaimUnitShapes = func() (map[string]claimUnitShape, error) { return map[string]claimUnitShape{}, nil } |
| 504 | wrapped := ClaimRootWithLaunch{Launch: f.root} |
| 505 | wrapped.Root.Stop.Coordinator.Kernel = claimUnitStopKernel{Units: f.root.Units} |
| 506 | assembly := &ClaimAssembly{} |
| 507 | if err := bindProductionClaimUnits(t.Context(), &wrapped, assembly); err != nil { |
| 508 | t.Fatal(err) |
| 509 | } |
| 510 | t.Cleanup(func() { assembly.Close(context.Background()) }) |
| 511 | dispatch := f.root.Units.(claimUnitDispatch) |
| 512 | plan14BoundOverrideRefusals(t, dispatch, host, unit, ctl) |
| 513 | if _, err := dispatch.forStart(t.Context(), host); err != nil { |
| 514 | t.Fatalf("ordinary authorized configuration: %#v", err) |
| 515 | } |
| 516 | } |
| 517 | var material *claimUnitMaterializer |
| 518 | if materialized { |
| 519 | ms, err := rootstate.Open(filepath.Join(dir, "unit-materializations"), true, false) |
| 520 | if err != nil { |
| 521 | t.Fatal(err) |
| 522 | } |
| 523 | defer ms.Close() |
| 524 | files := claimUnitFiles{dir: "/etc/systemd/system"} |
| 525 | material = &claimUnitMaterializer{idle: func(host string) error { return f.root.Units.Gone(t.Context(), host, "") }, state: ms, read: files.read, create: files.create, replace: files.replace} |
| 526 | binding, err := f.root.unitBindings.read(host) |
| 527 | if err != nil { |
| 528 | t.Fatal(err) |
| 529 | } |
| 530 | if migrated { |
| 531 | kernel := claimPredecessorKernel{ctl: ctl, group: claimPredecessorGroup, boot: claimUnitBoot} |
| 532 | material.predecessor, material.stopPredecessor = kernel.capture, kernel.stop |
| 533 | material.socketCapture, material.socketPark = captureProductionClaimUnitSocket, parkProductionClaimUnitSocket |
| 534 | if !tcpMode { |
| 535 | predecessor, err := kernel.capture(t.Context(), binding) |
| 536 | if err != nil { |
| 537 | t.Fatal(err) |
| 538 | } |
| 539 | predecessor.Group = "system.slice/unrelated.service" |
| 540 | if _, err := captureProductionClaimUnitSocket(t.Context(), binding.Shape, predecessor); err == nil { |
| 541 | t.Fatal("socket accepted a foreign predecessor cgroup") |
| 542 | } |
| 543 | } |
| 544 | } |
| 545 | if deferReplacement { |
| 546 | plan14ProductionStartupDefersReplacement(t, f.root, material, binding, ctl) |
| 547 | return |
| 548 | } |
| 549 | if err := material.install(t.Context(), binding); err != nil { |
| 550 | debug, _ := ctl(t.Context(), "show", unit, "-p", "ActiveState,InvocationID,ControlGroup,FragmentPath,DropInPaths") |
| 551 | t.Fatalf("migration/materialization: %v; unit: %s", err, debug) |
| 552 | } |
| 553 | if err := projectClaimBoundSocket(binding.Shape); err != nil { |
| 554 | t.Fatal(err) |
| 555 | } |
| 556 | if _, err := ctl(t.Context(), "daemon-reload"); err != nil { |
| 557 | t.Fatal(err) |
| 558 | } |
| 559 | } |
| 560 | if programMode { |
| 561 | base := f.root.Units.(instancectl.ClaimUnits) |
| 562 | base.BoundHost, base.BoundUnit = "", "" |
| 563 | binding, err := f.root.unitBindings.read(host) |
| 564 | if err != nil { |
| 565 | t.Fatal(err) |
| 566 | } |
| 567 | f.root.Units = claimUnitDispatch{base: base, bindings: f.root.unitBindings, shapes: func() (map[string]claimUnitShape, error) { return map[string]claimUnitShape{host: binding.Shape}, nil }, definition: func(s claimUnitShape) error { |
| 568 | if material != nil { |
| 569 | if err := material.verify(binding); err != nil { |
| 570 | return err |
| 571 | } |
| 572 | if err := claimEffectiveUnit(t.Context(), s, ctl); err != nil { |
| 573 | t.Logf("effective unit refused: %#v", err) |
| 574 | return err |
| 575 | } |
| 576 | return nil |
| 577 | } |
| 578 | // Dependencies and output are runner adapters; all isolation settings |
| 579 | // and the permanent no-reservation refusal remain in the unit. |
| 580 | raw, err := os.ReadFile(unitPath) |
| 581 | if err != nil { |
| 582 | return err |
| 583 | } |
| 584 | if string(raw) != strings.Join(lines, "\n") { |
| 585 | return os.ErrPermission |
| 586 | } |
| 587 | return nil |
| 588 | }} |
| 589 | } |
| 590 | if materialized { |
| 591 | plan14BoundOverrideRefusals(t, f.root.Units.(claimUnitDispatch), host, unit, ctl) |
| 592 | f.root.boundReady = func(ctx context.Context, host string, started instancectl.ClaimUnitStarted) error { |
| 593 | b, err := f.root.unitBindings.check(ctx, host, nil) |
| 594 | if err != nil { |
| 595 | return err |
| 596 | } |
| 597 | return claimBoundUnitReady(ctx, started, b.Shape) |
| 598 | } |
| 599 | } |
| 600 | f.root.Kernel = linuxClaimLaunchKernel{OptRoot: "/opt/carlos", CgroupRoot: "/sys/fs/cgroup", ProcRoot: "/proc"} |
| 601 | store := release.NewDirStore(filepath.Join(dir, "deployment")) |
| 602 | rec.Schema = 1 |
| 603 | plan14PutRecord(t, store, rec) |
| 604 | pub, priv, err := ed25519.GenerateKey(nil) |
| 605 | if err != nil { |
| 606 | t.Fatal(err) |
| 607 | } |
| 608 | runnerBody := body |
| 609 | if consoleBinary { |
| 610 | body = plan14ConsoleExecutable |
| 611 | } |
| 612 | if woodstar { |
| 613 | body = plan14WoodstarExecutable |
| 614 | } |
| 615 | for _, v := range []string{"v1", "v3"} { |
| 616 | if err := release.StageArtifact("/opt/carlos", rec.Account, v, rec.App, body); err != nil { |
| 617 | t.Fatal(err) |
| 618 | } |
| 619 | m := release.Manifest{App: rec.App, Version: v, Kind: "binary", Artifacts: map[string]release.Artifact{release.AdoptPlatform(): {}}} |
| 620 | if err := release.Ship(t.Context(), store, rec.Account, m, map[string][]byte{release.AdoptPlatform(): body}); err != nil { |
| 621 | t.Fatal(err) |
| 622 | } |
| 623 | if err := release.Promote(t.Context(), store, priv, rec.Account, rec.App, "canary/console", v); err != nil { |
| 624 | t.Fatal(err) |
| 625 | } |
| 626 | } |
| 627 | path := release.VersionPath("/opt/carlos", rec.Account, "v1", rec.App) |
| 628 | file, err := rootstate.OpenTrustedFile(path) |
| 629 | if err != nil { |
| 630 | t.Fatal(err) |
| 631 | } |
| 632 | id, err := rootstate.Identity(file) |
| 633 | file.Close() |
| 634 | if err != nil { |
| 635 | t.Fatal(err) |
| 636 | } |
| 637 | digest := sha256.Sum256(body) |
| 638 | f.req.Executable = path |
| 639 | f.req.Reservation.Executable = hostbox.LaunchFile{Device: id.Device, Inode: id.Inode} |
| 640 | f.req.Spec.Executable = path |
| 641 | f.req.Spec.Digest = hex.EncodeToString(digest[:]) |
| 642 | f.req.Spec.Args = []string{"--socket", filepath.Join(provision.InstanceRunDir(host), "app.sock"), "--db", provision.NestedDBPath("/data", host)} |
| 643 | f.req.Spec.Env = []string{"CARLOS_ADMIN_TOKEN=test", "PLAN14_TENANT_SOCKET=" + filepath.Join(provision.InstanceRunDir(host), "app.sock")} |
| 644 | if programMode { |
| 645 | f.req.Spec.Args, f.req.Spec.Env = program.Args, program.Env |
| 646 | f.req.Spec.Identity.Config = hex.EncodeToString(programSum[:]) |
| 647 | if materialized { |
| 648 | b, err := f.root.unitBindings.read(host) |
| 649 | if err != nil { |
| 650 | t.Fatal(err) |
| 651 | } |
| 652 | route := b.Shape.route() |
| 653 | out, err := f.root.Serve(t.Context(), f.peer, hostbox.RoleLaunchRequest{V: hostbox.RoleLaunchV, Op: hostbox.RoleLaunchInputs, Route: &route}) |
| 654 | if err != nil || out.Inputs == nil { |
| 655 | t.Fatal("bound admission inputs", err) |
| 656 | } |
| 657 | f.req.Spec.Args, f.req.Spec.Env = out.Inputs.Args, out.Inputs.Env |
| 658 | } |
| 659 | } |
| 660 | if len(runtimeUpdate) > 0 && !runtimeUpdate[0] { |
| 661 | f.req.Spec.Env = append(f.req.Spec.Env, "PLAN14_TENANT_FAIL=1") |
| 662 | f.root.ReadyBudget = 100 * time.Millisecond |
| 663 | } |
| 664 | inputs, err := f.req.Spec.Hash() |
| 665 | if err != nil { |
| 666 | t.Fatal(err) |
| 667 | } |
| 668 | f.req.Reservation.Inputs = inputs |
| 669 | f.claims.set(func(c *hostclaim.Claim) { c.Holder.WakeRetry.Attempt.LaunchInputs = inputs }) |
| 670 | f.root.generations = &rootGenerationSource{store: store, optDir: "/opt/carlos", pub: func() (ed25519.PublicKey, error) { return pub, nil }, open: func() (rootstate.Store, error) { return rootstate.Open(filepath.Join(dir, "generations"), true, false) }, openFile: rootstate.OpenTrustedFile, persistent: rootstate.PersistentIdentity} |
| 671 | var updateSaved rootUpdateRecord |
| 672 | var runtimeBook *rollout.RootFailures |
| 673 | if len(runtimeUpdate) > 0 { |
| 674 | state, err := rootstate.Open(filepath.Join(dir, "failures"), true, false) |
| 675 | if err != nil { |
| 676 | t.Fatal(err) |
| 677 | } |
| 678 | defer state.Close() |
| 679 | book, err := rollout.InitializeRootFailures(state, "box/en/1", nil) |
| 680 | if err != nil { |
| 681 | t.Fatal(err) |
| 682 | } |
| 683 | updateSaved = rootUpdateRecord{Schema: 1, Installation: "fixture-installation", Route: registry.Route{Host: host, Account: rec.Account, App: rec.App, Socket: filepath.Join(provision.InstanceRunDir(host), "app.sock")}, Consumed: &release.Consumed{Resolved: &release.ResolvedRequest{ID: 7, Target: "v1"}}} |
| 684 | runtimeBook = book |
| 685 | source := rootUpdateSource{state: state, book: book, ready: func(rt registry.Route) error { return rootUpdateReady(t.Context(), rt) }} |
| 686 | f.root.generations.beginUpdate = func(ctx context.Context, _ *rootGenerationSource, b release.GenerationBinding) (func(string, error, func() error) error, func() error, error) { |
| 687 | finish, err := source.beginRuntime(&updateSaved, func() error { raw, _ := json.Marshal(updateSaved); return state.Write("settings.json", raw) }, b.Version, b.Digest, b.Attempt) |
| 688 | return finish, func() error { return nil }, err |
| 689 | } |
| 690 | } |
| 691 | lock := &release.ReleaseExclusion{Open: func() (rootstate.Store, error) { return rootstate.Open(filepath.Join(dir, "release"), true, false) }} |
| 692 | checkSibling := func() {} |
| 693 | if programMode { |
| 694 | checkSibling = plan14BoundSibling(t, runnerBody, ctl, uid, gid, lock) |
| 695 | checkSibling() |
| 696 | } |
| 697 | if err := f.req.Validate(); err != nil { |
| 698 | t.Fatal("request", err) |
| 699 | } |
| 700 | if _, err := f.root.launchSpec(f.req); err != nil { |
| 701 | t.Fatal("launch spec", err) |
| 702 | } |
| 703 | if err := f.root.Kernel.Verify(f.req.Executable, f.req.Reservation.Executable); err != nil { |
| 704 | t.Fatal("kernel build", err) |
| 705 | } |
| 706 | var response hostbox.RoleLaunchResponse |
| 707 | err = lock.With(t.Context(), func(ctx context.Context) error { |
| 708 | var err error |
| 709 | response, err = f.root.Serve(ctx, f.peer, f.req) |
| 710 | return err |
| 711 | }) |
| 712 | if len(runtimeUpdate) > 0 && !runtimeUpdate[0] { |
| 713 | if err == nil || runtimeBook == nil || !runtimeBook.HeldDown(rec.Account, rec.App, "v1", host) || updateSaved.Build.State == nil || !updateSaved.Build.State.CompletedFailed || updateSaved.Build.State.Completed != 7 { |
| 714 | t.Fatal("installed failure lacked durable root refusal", err, updateSaved) |
| 715 | } |
| 716 | return |
| 717 | } |
| 718 | if err != nil { |
| 719 | logs, _ := exec.Command("journalctl", "-u", unit, "--no-pager", "-n", "80").CombinedOutput() |
| 720 | output, _ := os.ReadFile(provision.LogSocket(host, "run")) |
| 721 | t.Fatalf("production launch: %v\n%s\noutput: %s", err, logs, output) |
| 722 | } |
| 723 | if len(runtimeUpdate) > 0 && (updateSaved.Build.State == nil || updateSaved.Build.State.Completed != 7 || updateSaved.Build.State.CompletedFailed || updateSaved.Served["v1"].Invocation != response.Grant.UnitInvocationID) { |
| 724 | t.Fatal("root installed completion missing", updateSaved) |
| 725 | } |
| 726 | transport := &http.Transport{DialContext: func(ctx context.Context, _, _ string) (net.Conn, error) { |
| 727 | if tcpMode { |
| 728 | return (&net.Dialer{}).DialContext(ctx, "tcp", endpoint) |
| 729 | } |
| 730 | return (&net.Dialer{}).DialContext(ctx, "unix", socketEndpoint) |
| 731 | }} |
| 732 | defer transport.CloseIdleConnections() |
| 733 | client := &http.Client{Transport: transport, Timeout: 3 * time.Second} |
| 734 | resp, err := client.Get(healthURL) |
| 735 | if err != nil { |
| 736 | t.Fatal("installed tenant did not serve", err) |
| 737 | } |
| 738 | resp.Body.Close() |
| 739 | transport.CloseIdleConnections() |
| 740 | if resp.StatusCode != 200 { |
| 741 | t.Fatal(resp.Status) |
| 742 | } |
| 743 | if checkService != nil { |
| 744 | checkService(client) |
| 745 | } |
| 746 | if isolation || realConsole || woodstar { |
| 747 | if tcpMode { |
| 748 | siblingTransport := &http.Transport{DialContext: func(ctx context.Context, _, _ string) (net.Conn, error) { |
| 749 | return (&net.Dialer{}).DialContext(ctx, "unix", filepath.Join(provision.InstanceRunDir("sibling.example"), "app.sock")) |
| 750 | }} |
| 751 | siblingClient := &http.Client{Transport: siblingTransport, Timeout: 3 * time.Second} |
| 752 | response, err := siblingClient.Get("http://sibling/?dial=" + url.QueryEscape(endpoint)) |
| 753 | if err != nil { |
| 754 | t.Fatal("sibling isolation probe", err) |
| 755 | } |
| 756 | response.Body.Close() |
| 757 | siblingTransport.CloseIdleConnections() |
| 758 | if response.StatusCode != http.StatusForbidden { |
| 759 | t.Fatal("sibling reached bound TCP upstream", response.StatusCode) |
| 760 | } |
| 761 | } |
| 762 | for _, name := range func() []string { |
| 763 | if realConsole || woodstar { |
| 764 | return nil |
| 765 | } |
| 766 | return []string{"console", "sibling"} |
| 767 | }() { |
| 768 | listener, err := net.Listen("tcp", "127.0.0.1:0") |
| 769 | if err != nil { |
| 770 | t.Fatal(err) |
| 771 | } |
| 772 | conn, err := net.DialTimeout("tcp", listener.Addr().String(), time.Second) |
| 773 | if err != nil { |
| 774 | t.Fatal("control listener unavailable", err) |
| 775 | } |
| 776 | conn.Close() |
| 777 | for _, address := range []string{listener.Addr().String(), "[::ffff:127.0.0.1]:" + strings.Split(listener.Addr().String(), ":")[1]} { |
| 778 | response, err := client.Get("http://tenant/?dial=" + url.QueryEscape(address)) |
| 779 | if err != nil { |
| 780 | t.Fatal(err) |
| 781 | } |
| 782 | response.Body.Close() |
| 783 | if response.StatusCode != http.StatusForbidden { |
| 784 | t.Fatal("bound tenant reached", name, address, response.StatusCode) |
| 785 | } |
| 786 | } |
| 787 | listener.Close() |
| 788 | } |
| 789 | before, err := ctl(t.Context(), "show", "-p", "InvocationID", "--value", unit) |
| 790 | if err != nil { |
| 791 | t.Fatal(err) |
| 792 | } |
| 793 | if _, err := ctl(t.Context(), "restart", "carlos-edge.service"); err != nil { |
| 794 | t.Fatal(err) |
| 795 | } |
| 796 | after, err := ctl(t.Context(), "show", "-p", "InvocationID", "--value", unit) |
| 797 | if err != nil || before == "" || after != before { |
| 798 | t.Fatal("edge restart changed bound invocation", before, after, err) |
| 799 | } |
| 800 | response, err := client.Get(healthURL) |
| 801 | if err != nil { |
| 802 | t.Fatal("edge restart stopped bound service", err) |
| 803 | } |
| 804 | response.Body.Close() |
| 805 | if response.StatusCode != http.StatusOK { |
| 806 | t.Fatal(response.Status) |
| 807 | } |
| 808 | } |
| 809 | checkSibling() |
| 810 | if materialized { |
| 811 | b, err := f.root.unitBindings.read(host) |
| 812 | if err != nil { |
| 813 | t.Fatal(err) |
| 814 | } |
| 815 | unrelated := instancectl.ClaimUnitStarted{Unit: response.Grant.Unit, MainPID: response.Grant.PID, ControlGroup: instancectl.ClaimUnitControlGroup("unrelated.example")} |
| 816 | if b.Shape.Socket != "" { |
| 817 | unrelated.MainPID = os.Getpid() |
| 818 | } |
| 819 | if err := claimBoundUnitReady(t.Context(), unrelated, b.Shape); err == nil { |
| 820 | t.Fatal("readiness accepted unrelated process or group") |
| 821 | } |
| 822 | } |
| 823 | generations, err := rootstate.Open(filepath.Join(dir, "generations"), false, false) |
| 824 | if err != nil { |
| 825 | t.Fatal(err) |
| 826 | } |
| 827 | defer generations.Close() |
| 828 | ledger := &release.RootGenerations{Store: generations} |
| 829 | records, err := ledger.Reservations(t.Context()) |
| 830 | if err != nil || len(records) != 1 { |
| 831 | t.Fatal(records, err) |
| 832 | } |
| 833 | v1 := records[0] |
| 834 | if consoleBinary || woodstar { |
| 835 | running, err := os.ReadFile(fmt.Sprintf("/proc/%d/exe", v1.Process.TenantPID)) |
| 836 | if err != nil || sha256.Sum256(running) != sha256.Sum256(body) { |
| 837 | t.Fatal("running service is not the built executable", err) |
| 838 | } |
| 839 | } |
| 840 | if v1.Process.PID != response.Grant.PID || v1.Process.Containment != 1 || v1.Process.TenantPID == 0 { |
| 841 | t.Fatal("missing production process evidence", v1) |
| 842 | } |
| 843 | pending := v1.Binding |
| 844 | pending.Version = "v3" |
| 845 | pending.Attempt = "pending-v3" |
| 846 | pending.Invocation = "pending-v3" |
| 847 | pending.Path = release.VersionPath("/opt/carlos", rec.Account, "v3", rec.App) |
| 848 | pending.Tree = filepath.Dir(pending.Path) |
| 849 | for _, v := range []struct { |
| 850 | path string |
| 851 | id *rootstate.FileIdentity |
| 852 | }{{pending.Path, &pending.Executable}, {pending.Tree, &pending.TreeIdentity}} { |
| 853 | file, err := os.Open(v.path) |
| 854 | if err != nil { |
| 855 | t.Fatal(err) |
| 856 | } |
| 857 | *v.id, err = rootstate.Identity(file) |
| 858 | file.Close() |
| 859 | if err != nil { |
| 860 | t.Fatal(err) |
| 861 | } |
| 862 | } |
| 863 | if err := lock.With(t.Context(), func(ctx context.Context) error { _, err := ledger.Reserve(pending); return err }); err != nil { |
| 864 | t.Fatal(err) |
| 865 | } |
| 866 | retire := generationRetirement{guard: f.root.Guard, journal: journal, kernel: productionGenerationAbsence} |
| 867 | if err := lock.With(t.Context(), func(ctx context.Context) error { return retire.retire(ctx, ledger, v1) }); err != nil { |
| 868 | t.Fatal(err) |
| 869 | } |
| 870 | still, err := ledger.Read(v1.Binding) |
| 871 | if err != nil || still.Retirement != "" { |
| 872 | t.Fatal("live pin retired", err) |
| 873 | } |
| 874 | stop := f.req |
| 875 | stop.Op = hostbox.RoleLaunchStop |
| 876 | stop.Spec, stop.Executable, stop.MemoryMax = nil, "", 0 |
| 877 | if _, err := f.root.Serve(t.Context(), f.peer, stop); err != nil { |
| 878 | t.Fatal("production stop", err) |
| 879 | } |
| 880 | if programMode { |
| 881 | ctl(t.Context(), "start", unit) |
| 882 | time.Sleep(100 * time.Millisecond) |
| 883 | state, err := ctl(t.Context(), "show", "-p", "MainPID", "--value", unit) |
| 884 | if err != nil || state != "0" { |
| 885 | t.Fatal("unreserved restart executed", state, err) |
| 886 | } |
| 887 | ctl(t.Context(), "stop", unit) |
| 888 | next := *f.req.Reservation |
| 889 | next.ID, next.InvocationID = "attempt-2:main", "inv-2" |
| 890 | f.req.Reservation = &next |
| 891 | f.claims.set(func(c *hostclaim.Claim) { |
| 892 | c.Holder.WakeRetry.AttemptsReserved++ |
| 893 | c.Holder.WakeRetry.Attempt.ID = "attempt-2" |
| 894 | c.Holder.WakeRetry.Attempt.InvocationID = "inv-2" |
| 895 | }) |
| 896 | var restarted hostbox.RoleLaunchResponse |
| 897 | if err := lock.With(t.Context(), func(ctx context.Context) error { |
| 898 | var err error |
| 899 | restarted, err = f.root.Serve(ctx, f.peer, f.req) |
| 900 | return err |
| 901 | }); err != nil { |
| 902 | logs, _ := exec.Command("journalctl", "-u", unit, "--no-pager", "-n", "100").CombinedOutput() |
| 903 | state, _ := ctl(t.Context(), "show", "-p", "ActiveState,SubState,InvocationID,ControlGroup,ExecStart,Result", unit) |
| 904 | t.Fatalf("reserved restart: %v\n%s\n%s", err, state, logs) |
| 905 | } |
| 906 | if restarted.Grant.UnitInvocationID == response.Grant.UnitInvocationID { |
| 907 | t.Fatal("restart reused invocation") |
| 908 | } |
| 909 | resp, err := client.Get(healthURL) |
| 910 | if err != nil { |
| 911 | t.Fatal("restarted endpoint", err) |
| 912 | } |
| 913 | resp.Body.Close() |
| 914 | transport.CloseIdleConnections() |
| 915 | if resp.StatusCode != 200 { |
| 916 | t.Fatal(resp.Status) |
| 917 | } |
| 918 | if checkService != nil { |
| 919 | checkService(client) |
| 920 | } |
| 921 | stop.Reservation = &next |
| 922 | if _, err := f.root.Serve(t.Context(), f.peer, stop); err != nil { |
| 923 | t.Fatal("restarted stop", err) |
| 924 | } |
| 925 | } |
| 926 | checkSibling() |
| 927 | if err := journal.CloseWriters(t.Context(), f.target); err != nil { |
| 928 | t.Fatal(err) |
| 929 | } |
| 930 | if err := lock.With(t.Context(), func(ctx context.Context) error { return retire.retire(ctx, ledger, v1) }); err == nil { |
| 931 | t.Fatal("living controller released pin") |
| 932 | } |
| 933 | join() |
| 934 | raw, _ := json.Marshal(f.target) |
| 935 | if err := os.WriteFile(filepath.Join(dir, "target.json"), raw, 0600); err != nil { |
| 936 | t.Fatal(err) |
| 937 | } |
| 938 | // Isolate the census from unrelated runner daemons, as the generation suite |
| 939 | // does. The installed process and its controller were stopped and joined in |
| 940 | // the outer namespace first; the child uses the retained root records. |
| 941 | cmd := exec.CommandContext(t.Context(), "unshare", "--mount", "--pid", "--fork", "--mount-proc", binary, "-test.run=^TestPlan14InstalledClaimLaunchRootRunner$", "-test.v", "-test.timeout=2m") |
| 942 | cmd.Env = append(os.Environ(), "PLAN14_INSTALLED_RETIRE_CHILD="+dir) |
| 943 | if output, err := cmd.CombinedOutput(); err != nil { |
| 944 | t.Fatalf("exact pin release: %v\n%s", err, output) |
| 945 | } else { |
| 946 | t.Logf("%s", output) |
| 947 | } |
| 948 | checkSibling() |
| 949 | if material != nil { |
| 950 | retireUnit := func() error { |
| 951 | return retireClaimBoundUnit(t.Context(), f.target, f.root.unitBindings, *material, f.root.Units.Gone) |
| 952 | } |
| 953 | verifyCombined := func() {} |
| 954 | if combined { |
| 955 | retireUnit, verifyCombined = plan14CombinedBoundRetirement(t, dir, f, material, startup) |
| 956 | } |
| 957 | original, err := os.ReadFile(unitPath) |
| 958 | if err != nil { |
| 959 | t.Fatal(err) |
| 960 | } |
| 961 | saved := unitPath + ".plan14-original" |
| 962 | if _, err := os.Lstat(saved); !os.IsNotExist(err) { |
| 963 | t.Fatal("fixture predecessor exists", err) |
| 964 | } |
| 965 | if err := os.Rename(unitPath, saved); err != nil { |
| 966 | t.Fatal(err) |
| 967 | } |
| 968 | t.Cleanup(func() { os.Remove(saved) }) |
| 969 | files := claimUnitFiles{dir: filepath.Dir(unitPath)} |
| 970 | if err := files.create(unit, original); err != nil { |
| 971 | t.Fatal(err) |
| 972 | } |
| 973 | if err := retireUnit(); err == nil { |
| 974 | t.Fatal("same-byte successor retired") |
| 975 | } |
| 976 | after, err := os.ReadFile(unitPath) |
| 977 | if err != nil || string(after) != string(original) { |
| 978 | t.Fatal("successor changed", err) |
| 979 | } |
| 980 | if err := os.Remove(unitPath); err != nil { |
| 981 | t.Fatal(err) |
| 982 | } |
| 983 | if err := os.Rename(saved, unitPath); err != nil { |
| 984 | t.Fatal(err) |
| 985 | } |
| 986 | if err := retireUnit(); err != nil { |
| 987 | t.Fatal("unit retirement", err) |
| 988 | } |
| 989 | verifyCombined() |
| 990 | if _, err := ctl(t.Context(), "daemon-reload"); err != nil { |
| 991 | t.Fatal(err) |
| 992 | } |
| 993 | if _, err := ctl(t.Context(), "start", unit); err == nil { |
| 994 | t.Fatal("retired unit accepted start") |
| 995 | } |
| 996 | binding, err := f.root.unitBindings.read(host) |
| 997 | if err != nil || !binding.Retired { |
| 998 | t.Fatal(binding, err) |
| 999 | } |
| 1000 | for _, leaf := range []string{"launch.json", "permit.json", "exec.sock"} { |
| 1001 | if _, err := os.Lstat(filepath.Join("/run/carlos-control/exe/.claim", provision.InstanceID(host), leaf)); !os.IsNotExist(err) { |
| 1002 | t.Fatal("retired environment remains", leaf, err) |
| 1003 | } |
| 1004 | } |
| 1005 | // The root installation/claim source remains a fixture here. The durable |
| 1006 | // file transition and old-owner refusal use the real production adapters. |
| 1007 | f.root.unitBindings.handover = material.handover |
| 1008 | old.Stamp++ |
| 1009 | successor, err := f.root.unitBindings.establish(t.Context(), host) |
| 1010 | if err != nil { |
| 1011 | t.Fatal("successor binding", err) |
| 1012 | } |
| 1013 | if err := material.verify(successor); err != nil { |
| 1014 | t.Fatal("successor definition", err) |
| 1015 | } |
| 1016 | if err := retireClaimBoundUnit(t.Context(), f.target, f.root.unitBindings, *material, f.root.Units.Gone); err == nil { |
| 1017 | t.Fatal("old owner retired successor") |
| 1018 | } |
| 1019 | checkSibling() |
| 1020 | } |
| 1021 | } |
| 1022 | |
| 1023 | func plan14BoundOverrideRefusals(t *testing.T, dispatch claimUnitDispatch, host, unit string, ctl func(context.Context, ...string) (string, error)) { |
| 1024 | t.Helper() |
| 1025 | dir := filepath.Join("/etc/systemd/system", unit+".d") |
| 1026 | if err := os.Mkdir(dir, 0755); err != nil { |
| 1027 | t.Fatal(err) |
| 1028 | } |
| 1029 | t.Cleanup(func() { os.RemoveAll(dir) }) |
| 1030 | for _, attack := range []struct{ name, body string }{ |
| 1031 | {"90-operator.conf", "[Service]\nExecStart=\nExecStart=/usr/bin/true\n"}, |
| 1032 | {"90-user.conf", "[Service]\nUser=root\n"}, |
| 1033 | {"10-carlos-capabilities.conf", "[Service]\nCapabilityBoundingSet=CAP_SYS_ADMIN\n"}, |
| 1034 | {"90-mount.conf", "[Service]\nBindPaths=/etc:/tenant-secrets\n"}, |
| 1035 | } { |
| 1036 | path := filepath.Join(dir, attack.name) |
| 1037 | if err := os.WriteFile(path, []byte(attack.body), 0644); err != nil { |
| 1038 | t.Fatal(err) |
| 1039 | } |
| 1040 | if _, err := ctl(t.Context(), "daemon-reload"); err != nil { |
| 1041 | t.Fatal(err) |
| 1042 | } |
| 1043 | if _, err := dispatch.forStart(t.Context(), host); err == nil { |
| 1044 | t.Fatal("effective drop-in accepted", attack.name) |
| 1045 | } |
| 1046 | if err := os.Remove(path); err != nil { |
| 1047 | t.Fatal(err) |
| 1048 | } |
| 1049 | } |
| 1050 | if _, err := ctl(t.Context(), "daemon-reload"); err != nil { |
| 1051 | t.Fatal(err) |
| 1052 | } |
| 1053 | } |
| 1054 | |
| 1055 | func plan14InstalledRetire(t *testing.T, dir string) { |
| 1056 | state, err := rootstate.Open(filepath.Join(dir, "fences"), false, false) |
| 1057 | if err != nil { |
| 1058 | t.Fatal(err) |
| 1059 | } |
| 1060 | defer state.Close() |
| 1061 | journal, err := openGenerationLaunchInventory(filepath.Join(dir, "launch.db")) |
| 1062 | if err != nil { |
| 1063 | t.Fatal(err) |
| 1064 | } |
| 1065 | defer journal.Close() |
| 1066 | generations, err := rootstate.Open(filepath.Join(dir, "generations"), false, false) |
| 1067 | if err != nil { |
| 1068 | t.Fatal(err) |
| 1069 | } |
| 1070 | defer generations.Close() |
| 1071 | ledger := &release.RootGenerations{Store: generations} |
| 1072 | records, err := ledger.Reservations(t.Context()) |
| 1073 | if err != nil || len(records) < 2 { |
| 1074 | t.Fatal(records, err) |
| 1075 | } |
| 1076 | var v1s []release.GenerationReservation |
| 1077 | var v3 release.GenerationReservation |
| 1078 | for _, r := range records { |
| 1079 | if r.Binding.Version == "v1" { |
| 1080 | v1s = append(v1s, r) |
| 1081 | } else { |
| 1082 | v3 = r |
| 1083 | } |
| 1084 | } |
| 1085 | if err := os.Mkdir(filepath.Join(dir, "retire-locks"), 0700); err != nil { |
| 1086 | t.Fatal(err) |
| 1087 | } |
| 1088 | verifier := generationRetirement{guard: hostguard.Guard{DataDir: "/data", LockDir: filepath.Join(dir, "retire-locks"), Root: hostguard.NewRootAuthority(state)}, journal: journal, kernel: productionGenerationAbsence} |
| 1089 | lock := &release.ReleaseExclusion{Open: func() (rootstate.Store, error) { return rootstate.Open(filepath.Join(dir, "release"), false, false) }} |
| 1090 | for _, v1 := range v1s { |
| 1091 | if err := lock.With(t.Context(), func(ctx context.Context) error { return verifier.retire(ctx, ledger, v1) }); err != nil { |
| 1092 | t.Fatal(err) |
| 1093 | } |
| 1094 | released, err := ledger.Read(v1.Binding) |
| 1095 | if err != nil || released.Retirement == "" { |
| 1096 | t.Fatal("exact pin not released", released, err) |
| 1097 | } |
| 1098 | } |
| 1099 | pending, err := ledger.Read(v3.Binding) |
| 1100 | if err != nil || pending != v3 { |
| 1101 | t.Fatal("pending counterexample changed", pending, err) |
| 1102 | } |
| 1103 | } |
| 1104 | |
amadan