| 1 | //go:build linux |
| 2 | |
| 3 | package instanceexec |
| 4 | |
| 5 | import ( |
| 6 | "bufio" |
| 7 | "errors" |
| 8 | "fmt" |
| 9 | "io" |
| 10 | "net" |
| 11 | "os" |
| 12 | "os/exec" |
| 13 | "os/signal" |
| 14 | "strconv" |
| 15 | "strings" |
| 16 | "syscall" |
| 17 | "testing" |
| 18 | "time" |
| 19 | |
| 20 | "golang.org/x/sys/unix" |
| 21 | ) |
| 22 | |
| 23 | // The init tests run the real runInit in a child process: this test |
| 24 | // binary, re-executed with ieHelperEnv naming a role. The "init" role |
| 25 | // runs runInit with the binary itself as the tenant, in the tenant role |
| 26 | // named by ieTenantEnv; every other role is a tenant. |
| 27 | const ( |
| 28 | ieHelperEnv = "CARLOS_INSTANCEEXEC_HELPER" |
| 29 | ieTenantEnv = "CARLOS_INSTANCEEXEC_TENANT" |
| 30 | ieScopeEnv = "CARLOS_INSTANCEEXEC_SCOPE" |
| 31 | ieArgEnv = "CARLOS_INSTANCEEXEC_ARG" |
| 32 | ) |
| 33 | |
| 34 | func TestMain(m *testing.M) { |
| 35 | if role := os.Getenv(ieHelperEnv); role != "" { |
| 36 | os.Exit(runHelper(role)) |
| 37 | } |
| 38 | os.Exit(m.Run()) |
| 39 | } |
| 40 | |
| 41 | // withRole is the environment with the helper role replaced, not |
| 42 | // appended: a Go child keeps the FIRST of duplicate keys (design §6.2). |
| 43 | func withRole(role string) []string { |
| 44 | env := []string{ieHelperEnv + "=" + role} |
| 45 | for _, kv := range os.Environ() { |
| 46 | if !strings.HasPrefix(kv, ieHelperEnv+"=") { |
| 47 | env = append(env, kv) |
| 48 | } |
| 49 | } |
| 50 | return env |
| 51 | } |
| 52 | |
| 53 | func runHelper(role string) int { |
| 54 | self, err := os.Executable() |
| 55 | if err != nil { |
| 56 | fmt.Println("executable:", err) |
| 57 | return 70 |
| 58 | } |
| 59 | switch role { |
| 60 | case "init": |
| 61 | env := withRole(os.Getenv(ieTenantEnv)) |
| 62 | scope := func() error { return nil } |
| 63 | if os.Getenv(ieScopeEnv) == "1" { |
| 64 | scope = ScopeAbstractUnix |
| 65 | } |
| 66 | err := runInit(self, []string{self}, env, scope) |
| 67 | fmt.Println("init:", err) |
| 68 | return 71 |
| 69 | case "no-handler": |
| 70 | // A tenant that never asks for SIGTERM, like most Go servers. |
| 71 | fmt.Printf("ready pid=%d ppid=%d\n", os.Getpid(), os.Getppid()) |
| 72 | time.Sleep(30 * time.Second) |
| 73 | return 0 |
| 74 | case "echo-signals": |
| 75 | // Reports every forwarded signal; leaves on SIGTERM. |
| 76 | got := make(chan os.Signal, 8) |
| 77 | signal.Notify(got, forwarded[:]...) |
| 78 | fmt.Println("ready") |
| 79 | for s := range got { |
| 80 | fmt.Println("got", s) |
| 81 | if s == syscall.SIGTERM { |
| 82 | return 0 |
| 83 | } |
| 84 | } |
| 85 | return 0 |
| 86 | case "exit": |
| 87 | n, _ := strconv.Atoi(os.Getenv(ieArgEnv)) |
| 88 | return n |
| 89 | case "kill-self": |
| 90 | _ = syscall.Kill(os.Getpid(), syscall.SIGKILL) |
| 91 | time.Sleep(30 * time.Second) |
| 92 | return 0 |
| 93 | case "orphan": |
| 94 | // Leave a grandchild behind: the intermediate starts it and exits |
| 95 | // at once, so it is reparented to the nearest reaper. |
| 96 | cmd := exec.Command(self) |
| 97 | cmd.Env = withRole("intermediate") |
| 98 | cmd.Stdout = os.Stdout |
| 99 | if err := cmd.Run(); err != nil { |
| 100 | fmt.Println("intermediate:", err) |
| 101 | return 72 |
| 102 | } |
| 103 | time.Sleep(30 * time.Second) |
| 104 | return 0 |
| 105 | case "intermediate": |
| 106 | cmd := exec.Command(self) |
| 107 | cmd.Env = withRole("short-lived") |
| 108 | if err := cmd.Start(); err != nil { |
| 109 | fmt.Println("grandchild:", err) |
| 110 | return 73 |
| 111 | } |
| 112 | fmt.Printf("grandchild %d\n", cmd.Process.Pid) |
| 113 | return 0 |
| 114 | case "short-lived": |
| 115 | time.Sleep(1500 * time.Millisecond) |
| 116 | return 0 |
| 117 | case "inherit": |
| 118 | cwd, _ := os.Getwd() |
| 119 | sid, _ := unix.Getsid(0) |
| 120 | fmt.Printf("cwd=%s arg=%s sid-is-pid=%t\n", cwd, os.Getenv(ieArgEnv), sid == os.Getpid()) |
| 121 | f := os.NewFile(3, "fd3") |
| 122 | if _, err := f.Write([]byte("fd3 reached\n")); err != nil { |
| 123 | fmt.Println("fd3:", err) |
| 124 | } |
| 125 | return 0 |
| 126 | case "dial": |
| 127 | c, err := net.Dial("unix", os.Getenv(ieArgEnv)) |
| 128 | switch { |
| 129 | case err == nil: |
| 130 | c.Close() |
| 131 | fmt.Println("dial: connected") |
| 132 | case errors.Is(unwrapErrno(err), syscall.EPERM): |
| 133 | fmt.Println("dial: EPERM") |
| 134 | default: |
| 135 | fmt.Println("dial:", err) |
| 136 | } |
| 137 | return 0 |
| 138 | } |
| 139 | fmt.Println("unknown helper role", role) |
| 140 | return 74 |
| 141 | } |
| 142 | |
| 143 | // startInit starts the init helper with the given tenant role and extra |
| 144 | // environment, and returns it with a line reader on its stdout (which is |
| 145 | // the tenant's stdout too: the init passes fd 1 through). |
| 146 | func startInit(t *testing.T, tenant string, attr *syscall.SysProcAttr, extra ...string) (*exec.Cmd, *bufio.Reader) { |
| 147 | t.Helper() |
| 148 | self, err := os.Executable() |
| 149 | if err != nil { |
| 150 | t.Fatal(err) |
| 151 | } |
| 152 | cmd := exec.Command(self) |
| 153 | cmd.Env = append(os.Environ(), append([]string{ieHelperEnv + "=init", ieTenantEnv + "=" + tenant}, extra...)...) |
| 154 | cmd.Stderr = os.Stderr |
| 155 | cmd.SysProcAttr = attr |
| 156 | out, err := cmd.StdoutPipe() |
| 157 | if err != nil { |
| 158 | t.Fatal(err) |
| 159 | } |
| 160 | if err := cmd.Start(); err != nil { |
| 161 | if attr != nil { |
| 162 | t.Skipf("no unprivileged user and PID namespace here: %v", err) |
| 163 | } |
| 164 | t.Fatal(err) |
| 165 | } |
| 166 | t.Cleanup(func() { |
| 167 | // Through the init first, so the tenant stops with it outside a |
| 168 | // PID namespace too; the SIGKILL is for a test that failed early. |
| 169 | _ = cmd.Process.Signal(syscall.SIGTERM) |
| 170 | time.Sleep(100 * time.Millisecond) |
| 171 | _ = cmd.Process.Kill() |
| 172 | _ = cmd.Wait() |
| 173 | }) |
| 174 | return cmd, bufio.NewReader(out) |
| 175 | } |
| 176 | |
| 177 | // readLine reads one line within d, failing the test on a timeout. |
| 178 | func readLine(t *testing.T, r *bufio.Reader, d time.Duration) string { |
| 179 | t.Helper() |
| 180 | got := make(chan string, 1) |
| 181 | go func() { |
| 182 | line, err := r.ReadString('\n') |
| 183 | if err != nil && line == "" { |
| 184 | line = "read: " + err.Error() |
| 185 | } |
| 186 | got <- strings.TrimSpace(line) |
| 187 | }() |
| 188 | select { |
| 189 | case line := <-got: |
| 190 | return line |
| 191 | case <-time.After(d): |
| 192 | t.Fatalf("no line from the helper within %v", d) |
| 193 | return "" |
| 194 | } |
| 195 | } |
| 196 | |
| 197 | // waitExit waits up to d for cmd to exit and returns its exit code, or -1 |
| 198 | // with the signal if it died by one. |
| 199 | func waitExit(t *testing.T, cmd *exec.Cmd, d time.Duration) (int, syscall.Signal) { |
| 200 | t.Helper() |
| 201 | done := make(chan error, 1) |
| 202 | go func() { done <- cmd.Wait() }() |
| 203 | select { |
| 204 | case <-done: |
| 205 | case <-time.After(d): |
| 206 | t.Fatalf("helper still running %v after the signal", d) |
| 207 | } |
| 208 | ws := cmd.ProcessState.Sys().(syscall.WaitStatus) |
| 209 | if ws.Signaled() { |
| 210 | return -1, ws.Signal() |
| 211 | } |
| 212 | return ws.ExitStatus(), 0 |
| 213 | } |
| 214 | |
| 215 | func TestInitExitStatusMapping(t *testing.T) { |
| 216 | for _, c := range []struct { |
| 217 | ws syscall.WaitStatus |
| 218 | want int |
| 219 | }{ |
| 220 | {0, 0}, |
| 221 | {1 << 8, 1}, |
| 222 | {2 << 8, 2}, |
| 223 | {255 << 8, 255}, |
| 224 | {syscall.WaitStatus(syscall.SIGTERM), 143}, |
| 225 | {syscall.WaitStatus(syscall.SIGKILL), 137}, |
| 226 | {syscall.WaitStatus(syscall.SIGINT), 130}, |
| 227 | {syscall.WaitStatus(syscall.SIGSEGV) | 0x80, 139}, // core dumped |
| 228 | } { |
| 229 | if got := initExitCode(c.ws); got != c.want { |
| 230 | t.Errorf("initExitCode(%#x) = %d, want %d", uint32(c.ws), got, c.want) |
| 231 | } |
| 232 | } |
| 233 | } |
| 234 | |
| 235 | // End to end through the init, outside any namespace: the tenant's own |
| 236 | // exit code passes through, and a death by signal is 128+n. |
| 237 | func TestInitExitsWithTheTenantsStatus(t *testing.T) { |
| 238 | cmd, _ := startInit(t, "exit", nil, ieArgEnv+"=7") |
| 239 | if code, sig := waitExit(t, cmd, 10*time.Second); code != 7 { |
| 240 | t.Errorf("init exit = %d (signal %v), want 7", code, sig) |
| 241 | } |
| 242 | cmd, _ = startInit(t, "kill-self", nil) |
| 243 | if code, sig := waitExit(t, cmd, 10*time.Second); code != 128+int(syscall.SIGKILL) { |
| 244 | t.Errorf("init exit = %d (signal %v), want 137", code, sig) |
| 245 | } |
| 246 | } |
| 247 | |
| 248 | // Every forwarded signal reaches the tenant. |
| 249 | func TestInitForwardsEachSignal(t *testing.T) { |
| 250 | cmd, out := startInit(t, "echo-signals", nil) |
| 251 | if line := readLine(t, out, 10*time.Second); line != "ready" { |
| 252 | t.Fatalf("tenant said %q", line) |
| 253 | } |
| 254 | // SIGTERM last: the tenant leaves on it. |
| 255 | for _, sig := range []syscall.Signal{syscall.SIGHUP, syscall.SIGINT, syscall.SIGQUIT, syscall.SIGUSR1, syscall.SIGUSR2, syscall.SIGTERM} { |
| 256 | if err := cmd.Process.Signal(sig); err != nil { |
| 257 | t.Fatal(err) |
| 258 | } |
| 259 | if line := readLine(t, out, 5*time.Second); line != "got "+sig.String() { |
| 260 | t.Errorf("sent %v, tenant said %q", sig, line) |
| 261 | } |
| 262 | } |
| 263 | if code, sig := waitExit(t, cmd, 5*time.Second); code != 0 { |
| 264 | t.Errorf("init exit = %d (signal %v), want the tenant's 0", code, sig) |
| 265 | } |
| 266 | } |
| 267 | |
| 268 | // A tenant with no handlers acts on a forwarded signal by its default |
| 269 | // disposition, as the main process did before PrivatePIDs: it dies by |
| 270 | // the signal, and the init reports 128+n. |
| 271 | func TestInitStopsATenantWithNoHandler(t *testing.T) { |
| 272 | for _, sig := range []syscall.Signal{syscall.SIGTERM, syscall.SIGINT, syscall.SIGHUP} { |
| 273 | t.Run(sig.String(), func(t *testing.T) { |
| 274 | cmd, out := startInit(t, "no-handler", nil) |
| 275 | if line := readLine(t, out, 10*time.Second); !strings.HasPrefix(line, "ready ") { |
| 276 | t.Fatalf("tenant said %q", line) |
| 277 | } |
| 278 | if err := cmd.Process.Signal(sig); err != nil { |
| 279 | t.Fatal(err) |
| 280 | } |
| 281 | if code, got := waitExit(t, cmd, 5*time.Second); code != 128+int(sig) { |
| 282 | t.Errorf("init exit = %d (signal %v), want %d", code, got, 128+int(sig)) |
| 283 | } |
| 284 | }) |
| 285 | } |
| 286 | } |
| 287 | |
| 288 | // The tenant gets the init's cwd, the launch env, a session of its own |
| 289 | // (as the unit's main process had) and any inherited fd beyond stdio. |
| 290 | func TestInitPassesTheTenantItsEnvironment(t *testing.T) { |
| 291 | r, w, err := os.Pipe() |
| 292 | if err != nil { |
| 293 | t.Fatal(err) |
| 294 | } |
| 295 | defer r.Close() |
| 296 | self, _ := os.Executable() |
| 297 | cmd := exec.Command(self) |
| 298 | cmd.Dir = t.TempDir() |
| 299 | cmd.Env = append(os.Environ(), ieHelperEnv+"=init", ieTenantEnv+"=inherit", ieArgEnv+"=launch-value") |
| 300 | cmd.ExtraFiles = []*os.File{w} |
| 301 | var stdout strings.Builder |
| 302 | cmd.Stdout = &stdout |
| 303 | cmd.Stderr = os.Stderr |
| 304 | if err := cmd.Run(); err != nil { |
| 305 | t.Fatalf("init: %v; output %q", err, stdout.String()) |
| 306 | } |
| 307 | w.Close() |
| 308 | want := fmt.Sprintf("cwd=%s arg=launch-value sid-is-pid=true", cmd.Dir) |
| 309 | if got := strings.TrimSpace(stdout.String()); got != want { |
| 310 | t.Errorf("tenant said %q, want %q", got, want) |
| 311 | } |
| 312 | b, _ := io.ReadAll(r) |
| 313 | if string(b) != "fd3 reached\n" { |
| 314 | t.Errorf("fd 3 carried %q", b) |
| 315 | } |
| 316 | } |
| 317 | |
| 318 | // A process orphaned inside the namespace is the init's to reap. The |
| 319 | // grandchild must be reparented to the init (it is a subreaper, and PID 1 |
| 320 | // in production) and must not stay a zombie after it exits. |
| 321 | func TestInitReapsOrphans(t *testing.T) { |
| 322 | cmd, out := startInit(t, "orphan", nil) |
| 323 | line := readLine(t, out, 10*time.Second) |
| 324 | var gpid int |
| 325 | if _, err := fmt.Sscanf(line, "grandchild %d", &gpid); err != nil { |
| 326 | t.Fatalf("tenant said %q", line) |
| 327 | } |
| 328 | initPid := cmd.Process.Pid |
| 329 | adopted := false |
| 330 | deadline := time.Now().Add(10 * time.Second) |
| 331 | for time.Now().Before(deadline) { |
| 332 | state, ppid, ok := procStat(gpid) |
| 333 | if !ok { |
| 334 | break // reaped |
| 335 | } |
| 336 | if ppid == initPid { |
| 337 | adopted = true |
| 338 | } |
| 339 | if state == "Z" && time.Now().After(deadline.Add(-7*time.Second)) { |
| 340 | t.Fatalf("grandchild %d is a zombie of %d and nobody reaps it", gpid, ppid) |
| 341 | } |
| 342 | time.Sleep(20 * time.Millisecond) |
| 343 | } |
| 344 | if _, _, ok := procStat(gpid); ok { |
| 345 | t.Fatalf("grandchild %d still present", gpid) |
| 346 | } |
| 347 | if !adopted { |
| 348 | t.Fatalf("grandchild %d was never the init's (%d) child", gpid, initPid) |
| 349 | } |
| 350 | // The tenant itself is still running: reaping an orphan is not the |
| 351 | // tenant's exit. |
| 352 | if err := cmd.Process.Signal(syscall.Signal(0)); err != nil { |
| 353 | t.Fatalf("init gone after reaping an orphan: %v", err) |
| 354 | } |
| 355 | } |
| 356 | |
| 357 | // procStat returns the state and parent of pid from /proc/<pid>/stat. |
| 358 | func procStat(pid int) (state string, ppid int, ok bool) { |
| 359 | b, err := os.ReadFile(fmt.Sprintf("/proc/%d/stat", pid)) |
| 360 | if err != nil { |
| 361 | return "", 0, false |
| 362 | } |
| 363 | s := string(b) |
| 364 | i := strings.LastIndexByte(s, ')') |
| 365 | if i < 0 { |
| 366 | return "", 0, false |
| 367 | } |
| 368 | f := strings.Fields(s[i+1:]) |
| 369 | if len(f) < 2 { |
| 370 | return "", 0, false |
| 371 | } |
| 372 | ppid, _ = strconv.Atoi(f[1]) |
| 373 | return f[0], ppid, true |
| 374 | } |
| 375 | |
| 376 | // The Landlock scope applied by the init holds in the tenant: the scope is |
| 377 | // on the thread that forks, and a domain is inherited across fork and |
| 378 | // exec. The unscoped run is the control, proving the dial can succeed. |
| 379 | func TestInitTenantKeepsTheAbstractSocketScope(t *testing.T) { |
| 380 | abi, _, e := unix.Syscall(unix.SYS_LANDLOCK_CREATE_RULESET, 0, 0, unix.LANDLOCK_CREATE_RULESET_VERSION) |
| 381 | if e != 0 || abi < landlockScopeABI { |
| 382 | t.Skipf("no Landlock scope on this kernel (abi %d, %v)", abi, e) |
| 383 | } |
| 384 | name := "@carlos-init-test-" + randomName() |
| 385 | ln, err := net.Listen("unix", name) |
| 386 | if err != nil { |
| 387 | t.Skipf("abstract listen: %v", err) |
| 388 | } |
| 389 | defer ln.Close() |
| 390 | go func() { |
| 391 | for { |
| 392 | c, err := ln.Accept() |
| 393 | if err != nil { |
| 394 | return |
| 395 | } |
| 396 | c.Close() |
| 397 | } |
| 398 | }() |
| 399 | for scope, want := range map[string]string{"0": "dial: connected", "1": "dial: EPERM"} { |
| 400 | cmd, out := startInit(t, "dial", nil, ieArgEnv+"="+name, ieScopeEnv+"="+scope) |
| 401 | if got := readLine(t, out, 10*time.Second); got != want { |
| 402 | t.Errorf("scope=%s: tenant said %q, want %q", scope, got, want) |
| 403 | } |
| 404 | if code, sig := waitExit(t, cmd, 10*time.Second); code != 0 { |
| 405 | t.Errorf("scope=%s: init exit = %d (signal %v)", scope, code, sig) |
| 406 | } |
| 407 | } |
| 408 | } |
| 409 | |
| 410 | // The finding itself (#126, dev fleet): in a PID namespace, as under |
| 411 | // PrivatePIDs=yes, the first process is init and has no default signal |
| 412 | // actions. A Go tenant with no SIGTERM handler in that slot does not die |
| 413 | // by SIGTERM. Through the init it is the init's child and does, promptly. Needs an |
| 414 | // unprivileged user namespace; skipped where there is none. |
| 415 | func TestInitStopsATenantWithNoHandlerInAPIDNamespace(t *testing.T) { |
| 416 | userPIDNS := func() *syscall.SysProcAttr { |
| 417 | return &syscall.SysProcAttr{ |
| 418 | Cloneflags: syscall.CLONE_NEWUSER | syscall.CLONE_NEWPID, |
| 419 | UidMappings: []syscall.SysProcIDMap{{ContainerID: os.Getuid(), HostID: os.Getuid(), Size: 1}}, |
| 420 | GidMappings: []syscall.SysProcIDMap{{ContainerID: os.Getgid(), HostID: os.Getgid(), Size: 1}}, |
| 421 | } |
| 422 | } |
| 423 | |
| 424 | t.Run("tenant as PID 1", func(t *testing.T) { |
| 425 | self, _ := os.Executable() |
| 426 | cmd := exec.Command(self) |
| 427 | cmd.Env = append(os.Environ(), ieHelperEnv+"=no-handler") |
| 428 | cmd.SysProcAttr = userPIDNS() |
| 429 | cmd.Stderr = io.Discard // the runtime's own complaint, if any |
| 430 | out, err := cmd.StdoutPipe() |
| 431 | if err != nil { |
| 432 | t.Fatal(err) |
| 433 | } |
| 434 | if err := cmd.Start(); err != nil { |
| 435 | t.Skipf("no unprivileged user and PID namespace here: %v", err) |
| 436 | } |
| 437 | defer func() { _ = cmd.Process.Kill(); _ = cmd.Wait() }() |
| 438 | if line := readLine(t, bufio.NewReader(out), 10*time.Second); !strings.HasPrefix(line, "ready pid=1 ") { |
| 439 | t.Fatalf("tenant said %q, want it at PID 1", line) |
| 440 | } |
| 441 | _ = cmd.Process.Signal(syscall.SIGTERM) |
| 442 | done := make(chan struct{}) |
| 443 | go func() { _ = cmd.Wait(); close(done) }() |
| 444 | select { |
| 445 | case <-done: |
| 446 | ws := cmd.ProcessState.Sys().(syscall.WaitStatus) |
| 447 | if ws.Signaled() && ws.Signal() == syscall.SIGTERM { |
| 448 | t.Skip("this kernel delivered SIGTERM's default action to a namespace init") |
| 449 | } |
| 450 | t.Logf("as PID 1 the tenant did not die by SIGTERM: %v (the bug the init fixes)", cmd.ProcessState) |
| 451 | case <-time.After(2 * time.Second): |
| 452 | t.Logf("as PID 1 the tenant ignored SIGTERM (the bug the init fixes)") |
| 453 | } |
| 454 | }) |
| 455 | |
| 456 | t.Run("through the init", func(t *testing.T) { |
| 457 | cmd, out := startInit(t, "no-handler", userPIDNS()) |
| 458 | // Not PID 2 as such: the init's own threads take PIDs in the |
| 459 | // namespace too. What matters is that the tenant is not PID 1. |
| 460 | line := readLine(t, out, 10*time.Second) |
| 461 | var pid, ppid int |
| 462 | if _, err := fmt.Sscanf(line, "ready pid=%d ppid=%d", &pid, &ppid); err != nil || pid == 1 || ppid != 1 { |
| 463 | t.Fatalf("tenant said %q, want a child of the init at PID 1", line) |
| 464 | } |
| 465 | start := time.Now() |
| 466 | if err := cmd.Process.Signal(syscall.SIGTERM); err != nil { |
| 467 | t.Fatal(err) |
| 468 | } |
| 469 | code, sig := waitExit(t, cmd, 5*time.Second) |
| 470 | if code != 128+int(syscall.SIGTERM) { |
| 471 | t.Errorf("init exit = %d (signal %v), want 143", code, sig) |
| 472 | } |
| 473 | t.Logf("stopped in %v", time.Since(start)) |
| 474 | }) |
| 475 | } |
| 476 | |
amadan