| 1 | package caps |
| 2 | |
| 3 | import ( |
| 4 | "bytes" |
| 5 | "errors" |
| 6 | "fmt" |
| 7 | "os" |
| 8 | "path/filepath" |
| 9 | "strings" |
| 10 | "testing" |
| 11 | ) |
| 12 | |
| 13 | // fakeSystemd is a tiny in-memory systemd: it RECORDS every state-changing |
| 14 | // verb and answers the query seam from the state those verbs left behind. The |
| 15 | // state tracking is what lets "converge twice, the second pass changes |
| 16 | // nothing" be asserted against a DAEMON route — the interesting case — rather |
| 17 | // than only against a route with no unit state to repair. |
| 18 | type fakeSystemd struct { |
| 19 | calls []string |
| 20 | active map[string]bool |
| 21 | enabled map[string]bool |
| 22 | fail map[string]error // "verb unit" -> error |
| 23 | } |
| 24 | |
| 25 | func newFakeSystemd() *fakeSystemd { |
| 26 | return &fakeSystemd{active: map[string]bool{}, enabled: map[string]bool{}, fail: map[string]error{}} |
| 27 | } |
| 28 | |
| 29 | func (f *fakeSystemd) run(args ...string) error { |
| 30 | call := strings.Join(args, " ") |
| 31 | f.calls = append(f.calls, call) |
| 32 | if err := f.fail[call]; err != nil { |
| 33 | return err |
| 34 | } |
| 35 | if len(args) < 2 { |
| 36 | return nil |
| 37 | } |
| 38 | switch args[0] { |
| 39 | case "start", "restart": |
| 40 | f.active[args[1]] = true |
| 41 | case "stop": |
| 42 | f.active[args[1]] = false |
| 43 | case "enable": |
| 44 | f.enabled[args[1]] = true |
| 45 | case "disable": |
| 46 | f.enabled[args[1]] = false |
| 47 | } |
| 48 | return nil |
| 49 | } |
| 50 | |
| 51 | func (f *fakeSystemd) query(args ...string) bool { |
| 52 | if len(args) < 2 { |
| 53 | return false |
| 54 | } |
| 55 | switch args[0] { |
| 56 | case "is-active": |
| 57 | return f.active[args[1]] |
| 58 | case "is-enabled": |
| 59 | return f.enabled[args[1]] |
| 60 | } |
| 61 | return false |
| 62 | } |
| 63 | |
| 64 | func (f *fakeSystemd) reset() { f.calls = nil } |
| 65 | |
| 66 | // testOps wires every seam at a temp dir so nothing here touches systemd, |
| 67 | // /etc, /run or /var/lib. |
| 68 | func testOps(t *testing.T) (Ops, *fakeSystemd, *[]string, *[]string) { |
| 69 | t.Helper() |
| 70 | f := newFakeSystemd() |
| 71 | var logs []string |
| 72 | var users []string |
| 73 | o := Ops{ |
| 74 | EtcDir: t.TempDir(), |
| 75 | StateRoot: t.TempDir(), |
| 76 | RunDir: t.TempDir(), |
| 77 | Systemctl: f.run, |
| 78 | Query: f.query, |
| 79 | EnsureUser: func(name string) error { |
| 80 | users = append(users, name) |
| 81 | return nil |
| 82 | }, |
| 83 | Logf: func(format string, a ...any) { logs = append(logs, fmt.Sprintf(format, a...)) }, |
| 84 | } |
| 85 | return o, f, &logs, &users |
| 86 | } |
| 87 | |
| 88 | func keymailPlan(t *testing.T, g Grants) Plan { |
| 89 | t.Helper() |
| 90 | return Plan{Host: "keymail.dev", App: "keymail", Set: keymailSet(t), Grants: g} |
| 91 | } |
| 92 | |
| 93 | func readFile(t *testing.T, path string) []byte { |
| 94 | t.Helper() |
| 95 | b, err := os.ReadFile(path) |
| 96 | if err != nil { |
| 97 | t.Fatalf("read %s: %v", path, err) |
| 98 | } |
| 99 | return b |
| 100 | } |
| 101 | |
| 102 | func mustNotExist(t *testing.T, path string) { |
| 103 | t.Helper() |
| 104 | if _, err := os.Lstat(path); !os.IsNotExist(err) { |
| 105 | t.Fatalf("%s: want absent, got err=%v", path, err) |
| 106 | } |
| 107 | } |
| 108 | |
| 109 | func hasLog(logs []string, substr string) bool { |
| 110 | for _, l := range logs { |
| 111 | if strings.Contains(l, substr) { |
| 112 | return true |
| 113 | } |
| 114 | } |
| 115 | return false |
| 116 | } |
| 117 | |
| 118 | // indexOf returns the position of the first call equal to want, or -1. |
| 119 | func indexOf(calls []string, want string) int { |
| 120 | for i, c := range calls { |
| 121 | if c == want { |
| 122 | return i |
| 123 | } |
| 124 | } |
| 125 | return -1 |
| 126 | } |
| 127 | |
| 128 | // TestGrantAddedWritesTheDropInAndRestarts pins the whole happy path: both |
| 129 | // drop-ins are the golden bytes, exactly one daemon-reload, the socket |
| 130 | // sequence in its pinned order (a socket restart under a running service |
| 131 | // races into EADDRINUSE), and `enable` for BOTH units — bootstrap enables no |
| 132 | // tenant socket unit, so without it the ListenStream binds nothing after a |
| 133 | // reboot. |
| 134 | func TestGrantAddedWritesTheDropInAndRestarts(t *testing.T) { |
| 135 | o, f, _, users := testOps(t) |
| 136 | p := keymailPlan(t, keymailGrants()) |
| 137 | |
| 138 | res, err := Converge(o, p) |
| 139 | if err != nil { |
| 140 | t.Fatalf("Converge: %v", err) |
| 141 | } |
| 142 | if !res.Changed { |
| 143 | t.Fatal("Changed = false, want true on a first converge") |
| 144 | } |
| 145 | svcPath, sockPath := DropInPaths(o.EtcDir, p.Host) |
| 146 | if got, want := readFile(t, svcPath), readGolden(t, "keymail.service.conf"); !bytes.Equal(got, want) { |
| 147 | t.Fatalf("service drop-in:\n got: %q\n want: %q", got, want) |
| 148 | } |
| 149 | if got, want := readFile(t, sockPath), readGolden(t, "keymail.socket.conf"); !bytes.Equal(got, want) { |
| 150 | t.Fatalf("socket drop-in:\n got: %q\n want: %q", got, want) |
| 151 | } |
| 152 | if len(*users) != 1 || (*users)[0] != "carlos-keymail" { |
| 153 | t.Fatalf("EnsureUser calls = %v, want [carlos-keymail]", *users) |
| 154 | } |
| 155 | |
| 156 | reloads := 0 |
| 157 | for _, c := range f.calls { |
| 158 | if c == "daemon-reload" { |
| 159 | reloads++ |
| 160 | } |
| 161 | } |
| 162 | if reloads != 1 { |
| 163 | t.Fatalf("daemon-reload x%d, want exactly 1 (calls: %v)", reloads, f.calls) |
| 164 | } |
| 165 | |
| 166 | stop := indexOf(f.calls, "stop carlos-app@keymail.dev.service") |
| 167 | sock := indexOf(f.calls, "restart carlos-app@keymail.dev.socket") |
| 168 | start := indexOf(f.calls, "start carlos-app@keymail.dev.service") |
| 169 | if stop < 0 || sock < 0 || start < 0 || !(stop < sock && sock < start) { |
| 170 | t.Fatalf("socket ordering: stop=%d socket=%d start=%d, want stop < socket restart < start (calls: %v)", stop, sock, start, f.calls) |
| 171 | } |
| 172 | if indexOf(f.calls, "enable carlos-app@keymail.dev.service") < 0 { |
| 173 | t.Fatalf("no enable for the service unit (calls: %v)", f.calls) |
| 174 | } |
| 175 | if indexOf(f.calls, "enable carlos-app@keymail.dev.socket") < 0 { |
| 176 | t.Fatalf("no enable for the socket unit — the ListenStream would bind nothing after a reboot (calls: %v)", f.calls) |
| 177 | } |
| 178 | // The reload precedes EVERY unit verb: a restart against configuration |
| 179 | // systemd has not re-read is a pass that looks like it worked and changed |
| 180 | // nothing. |
| 181 | reload := indexOf(f.calls, "daemon-reload") |
| 182 | for i, c := range f.calls { |
| 183 | verb, _, _ := strings.Cut(c, " ") |
| 184 | switch verb { |
| 185 | case "restart", "start", "stop", "enable": |
| 186 | if i < reload { |
| 187 | t.Fatalf("%q ran before the daemon-reload: %v", c, f.calls) |
| 188 | } |
| 189 | } |
| 190 | } |
| 191 | } |
| 192 | |
| 193 | // A failed daemon-reload must not leave the new bytes on disk: the next pass |
| 194 | // would see no diff, conclude there is nothing to do, and the box would run |
| 195 | // the old configuration behind a drop-in that reads "converged" forever. |
| 196 | func TestAFailedReloadRollsTheBytesBackSoTheNextPassRetries(t *testing.T) { |
| 197 | o, f, _, _ := testOps(t) |
| 198 | p := keymailPlan(t, keymailGrants()) |
| 199 | f.fail["daemon-reload"] = errors.New("systemd is not running") |
| 200 | |
| 201 | res, err := Converge(o, p) |
| 202 | if err == nil { |
| 203 | t.Fatal("Converge swallowed a failed daemon-reload") |
| 204 | } |
| 205 | if res.Changed { |
| 206 | t.Fatal("Changed = true after the pass was rolled back") |
| 207 | } |
| 208 | svcPath, sockPath := DropInPaths(o.EtcDir, p.Host) |
| 209 | mustNotExist(t, svcPath) |
| 210 | mustNotExist(t, sockPath) |
| 211 | for _, c := range f.calls { |
| 212 | verb, _, _ := strings.Cut(c, " ") |
| 213 | if verb == "restart" || verb == "start" || verb == "stop" { |
| 214 | t.Fatalf("unit state changed after a failed reload: %v", f.calls) |
| 215 | } |
| 216 | } |
| 217 | |
| 218 | // Pass N+1, with a working systemd: the whole change is retried. |
| 219 | delete(f.fail, "daemon-reload") |
| 220 | f.reset() |
| 221 | res, err = Converge(o, p) |
| 222 | if err != nil { |
| 223 | t.Fatalf("Converge (retry): %v", err) |
| 224 | } |
| 225 | if !res.Changed { |
| 226 | t.Fatal("the retry pass found nothing to do — the rollback did not restore the diff") |
| 227 | } |
| 228 | if got, want := readFile(t, svcPath), readGolden(t, "keymail.service.conf"); !bytes.Equal(got, want) { |
| 229 | t.Fatalf("retry did not converge:\n got: %q\n want: %q", got, want) |
| 230 | } |
| 231 | if got, want := readFile(t, sockPath), readGolden(t, "keymail.socket.conf"); !bytes.Equal(got, want) { |
| 232 | t.Fatalf("retry did not converge the socket half:\n got: %q\n want: %q", got, want) |
| 233 | } |
| 234 | } |
| 235 | |
| 236 | // A half-written pass is the same divergence in miniature: one unit |
| 237 | // converged, the other not, and the disk claiming both are done. |
| 238 | func TestAPartialWriteRollsBothHalvesBackToPrev(t *testing.T) { |
| 239 | o, f, _, _ := testOps(t) |
| 240 | p := keymailPlan(t, keymailGrants()) |
| 241 | if _, err := Converge(o, p); err != nil { |
| 242 | t.Fatalf("Converge (first): %v", err) |
| 243 | } |
| 244 | svcPath, sockPath := DropInPaths(o.EtcDir, p.Host) |
| 245 | svcA, sockA := readFile(t, svcPath), readFile(t, sockPath) |
| 246 | f.reset() |
| 247 | |
| 248 | // Version B changes both halves, but the socket half cannot be written: |
| 249 | // its temp file lands in a directory that has been replaced by... itself, |
| 250 | // made unwritable. |
| 251 | if err := os.Chmod(filepath.Dir(sockPath), 0o500); err != nil { |
| 252 | t.Fatal(err) |
| 253 | } |
| 254 | t.Cleanup(func() { os.Chmod(filepath.Dir(sockPath), 0o755) }) |
| 255 | var b Set |
| 256 | if err := b.UnmarshalJSON([]byte(`{"v":1,"command":"forward","daemon":true,"runtime_dir":"keymail","host_role":true,"ports":[{"port":587,"name":"submission"}]}`)); err != nil { |
| 257 | t.Fatal(err) |
| 258 | } |
| 259 | p.Set = &b |
| 260 | p.Grants = Grants{Ports: map[int]bool{587: true}, HostRole: true} |
| 261 | |
| 262 | res, err := Converge(o, p) |
| 263 | if err == nil { |
| 264 | t.Skip("this environment allows writing into a 0500 directory (running as root?)") |
| 265 | } |
| 266 | if res.Changed { |
| 267 | t.Fatal("Changed = true after both halves were rolled back") |
| 268 | } |
| 269 | if got := readFile(t, svcPath); !bytes.Equal(got, svcA) { |
| 270 | t.Fatalf("the service half was left mid-change:\n got: %q\n want: %q", got, svcA) |
| 271 | } |
| 272 | if got := readFile(t, sockPath); !bytes.Equal(got, sockA) { |
| 273 | t.Fatalf("the socket half moved:\n got: %q\n want: %q", got, sockA) |
| 274 | } |
| 275 | for _, c := range f.calls { |
| 276 | verb, _, _ := strings.Cut(c, " ") |
| 277 | if verb == "restart" || verb == "start" || verb == "stop" { |
| 278 | t.Fatalf("unit state changed on a half-written pass: %v", f.calls) |
| 279 | } |
| 280 | } |
| 281 | } |
| 282 | |
| 283 | // Deviation #4's consequence, made visible: the socket unit is enabled for |
| 284 | // hibernating routes too (or the ListenStream binds nothing), and a granted |
| 285 | // public port is then a wake path that does not go through the activator's |
| 286 | // health gate. |
| 287 | func TestAGrantedPortOnAHibernatingRouteIsAdvised(t *testing.T) { |
| 288 | o, _, logs, _ := testOps(t) |
| 289 | var s Set |
| 290 | if err := s.UnmarshalJSON([]byte(`{"v":1,"command":"relay","ports":[{"port":25,"name":"mx"}]}`)); err != nil { |
| 291 | t.Fatal(err) |
| 292 | } |
| 293 | p := Plan{Host: "mx.example.com", App: "keymail", Set: &s, |
| 294 | Grants: Grants{Ports: map[int]bool{25: true}}, Hibernate: true} |
| 295 | |
| 296 | if _, err := Converge(o, p); err != nil { |
| 297 | t.Fatalf("Converge: %v", err) |
| 298 | } |
| 299 | if !hasLog(*logs, "port:25 is granted on a hibernating route") { |
| 300 | t.Fatalf("logs = %v, want the hibernating-port advisory", *logs) |
| 301 | } |
| 302 | if !hasLog(*logs, "bypassing the activator's health-gated wake") { |
| 303 | t.Fatalf("logs = %v, want the advisory to say what the risk is", *logs) |
| 304 | } |
| 305 | |
| 306 | // Not advised when the port is merely declared: nothing binds, so there is |
| 307 | // no wake path to warn about. Advice that fires on a healthy fleet is |
| 308 | // advice nobody reads. |
| 309 | o2, _, logs2, _ := testOps(t) |
| 310 | q := p |
| 311 | q.Host = "mx2.example.com" |
| 312 | q.Grants = Grants{} |
| 313 | if _, err := Converge(o2, q); err != nil { |
| 314 | t.Fatalf("Converge (ungranted): %v", err) |
| 315 | } |
| 316 | if hasLog(*logs2, "hibernating route") { |
| 317 | t.Fatalf("logs = %v, want no advisory for an ungranted port", *logs2) |
| 318 | } |
| 319 | } |
| 320 | |
| 321 | // An address-form ListenStream is the same claim: a hand-written drop-in |
| 322 | // cannot dodge the box-wide port check by spelling :25 differently. |
| 323 | func TestPortClaimsSeeAddressFormListenStreams(t *testing.T) { |
| 324 | for _, form := range []string{"25", "0.0.0.0:25", "[::]:25", "10.0.0.4:25"} { |
| 325 | o, _, logs, _ := testOps(t) |
| 326 | sockPath := filepath.Join(o.EtcDir, SocketUnit("hand.example.com")+".d", DropInName) |
| 327 | if err := os.MkdirAll(filepath.Dir(sockPath), 0o755); err != nil { |
| 328 | t.Fatal(err) |
| 329 | } |
| 330 | if err := os.WriteFile(sockPath, []byte("[Socket]\nListenStream="+form+"\n"), 0o644); err != nil { |
| 331 | t.Fatal(err) |
| 332 | } |
| 333 | if _, err := Converge(o, keymailPlan(t, keymailGrants())); err == nil { |
| 334 | t.Fatalf("ListenStream=%s did not count as a claim on :25", form) |
| 335 | } |
| 336 | if !hasLog(*logs, "port 25 is already claimed by hand.example.com") { |
| 337 | t.Fatalf("ListenStream=%s: logs = %v", form, *logs) |
| 338 | } |
| 339 | } |
| 340 | // The template's own unix listener claims no TCP port at all. |
| 341 | o, _, _, _ := testOps(t) |
| 342 | sockPath := filepath.Join(o.EtcDir, SocketUnit("hand.example.com")+".d", DropInName) |
| 343 | if err := os.MkdirAll(filepath.Dir(sockPath), 0o755); err != nil { |
| 344 | t.Fatal(err) |
| 345 | } |
| 346 | if err := os.WriteFile(sockPath, []byte("[Socket]\nListenStream=/run/carlos/apps/hand.example.com.sock\n"), 0o644); err != nil { |
| 347 | t.Fatal(err) |
| 348 | } |
| 349 | if _, err := Converge(o, keymailPlan(t, keymailGrants())); err != nil { |
| 350 | t.Fatalf("a unix ListenStream was read as a TCP claim: %v", err) |
| 351 | } |
| 352 | } |
| 353 | |
| 354 | // TestGrantRevokedRewritesTheDropIn: a grant is not a promise. Revoking |
| 355 | // port:25 and host-role removes the socket file entirely and rewrites the |
| 356 | // service file without its User= lines — and the removal alone is enough to |
| 357 | // count as Changed. |
| 358 | func TestGrantRevokedRewritesTheDropIn(t *testing.T) { |
| 359 | o, f, _, _ := testOps(t) |
| 360 | p := keymailPlan(t, keymailGrants()) |
| 361 | if _, err := Converge(o, p); err != nil { |
| 362 | t.Fatalf("Converge (granted): %v", err) |
| 363 | } |
| 364 | f.reset() |
| 365 | |
| 366 | p.Grants = Grants{} |
| 367 | res, err := Converge(o, p) |
| 368 | if err != nil { |
| 369 | t.Fatalf("Converge (revoked): %v", err) |
| 370 | } |
| 371 | if !res.Changed { |
| 372 | t.Fatal("Changed = false after a revoke, want true") |
| 373 | } |
| 374 | svcPath, sockPath := DropInPaths(o.EtcDir, p.Host) |
| 375 | mustNotExist(t, sockPath) |
| 376 | svc := string(readFile(t, svcPath)) |
| 377 | for _, gone := range []string{"DynamicUser=no", "User=carlos-keymail", "Group=carlos-keymail", "IPAddressAllow"} { |
| 378 | if strings.Contains(svc, gone) { |
| 379 | t.Fatalf("service drop-in still holds %q after the revoke:\n%s", gone, svc) |
| 380 | } |
| 381 | } |
| 382 | for _, kept := range []string{"ExecStart=/opt/carlos/live/%i relay", "RuntimeDirectory=keymail", "TasksMax=1024"} { |
| 383 | if !strings.Contains(svc, kept) { |
| 384 | t.Fatalf("service drop-in lost the self-service %q:\n%s", kept, svc) |
| 385 | } |
| 386 | } |
| 387 | reloads := 0 |
| 388 | for _, c := range f.calls { |
| 389 | if c == "daemon-reload" { |
| 390 | reloads++ |
| 391 | } |
| 392 | } |
| 393 | if reloads != 1 { |
| 394 | t.Fatalf("daemon-reload x%d, want exactly 1 (calls: %v)", reloads, f.calls) |
| 395 | } |
| 396 | if res.Prev.Socket == nil || !res.Prev.SocketHad { |
| 397 | t.Fatal("Prev did not capture the socket drop-in that was removed") |
| 398 | } |
| 399 | // A grant is not a promise (spec §7): the socket unit must be restarted |
| 400 | // so systemd actually drops :25. Removing the file and reloading is not |
| 401 | // enough — the running socket unit is still holding the listener. |
| 402 | stop := indexOf(f.calls, "stop carlos-app@keymail.dev.service") |
| 403 | sock := indexOf(f.calls, "restart carlos-app@keymail.dev.socket") |
| 404 | start := indexOf(f.calls, "start carlos-app@keymail.dev.service") |
| 405 | if stop < 0 || sock < 0 || start < 0 || !(stop < sock && sock < start) { |
| 406 | t.Fatalf("a revoked port did not go through the socket sequence: %v", f.calls) |
| 407 | } |
| 408 | } |
| 409 | |
| 410 | // TestUnreadableManifestLeavesDropInsAlone is the three-state rule, and the |
| 411 | // reason it exists: a bucket hiccup must not strip :25 off a running mail |
| 412 | // server. Nil Set = leave-alone, not removal. |
| 413 | func TestUnreadableManifestLeavesDropInsAlone(t *testing.T) { |
| 414 | o, f, logs, _ := testOps(t) |
| 415 | p := keymailPlan(t, keymailGrants()) |
| 416 | if _, err := Converge(o, p); err != nil { |
| 417 | t.Fatalf("Converge (staged): %v", err) |
| 418 | } |
| 419 | svcPath, sockPath := DropInPaths(o.EtcDir, p.Host) |
| 420 | svcBefore, sockBefore := readFile(t, svcPath), readFile(t, sockPath) |
| 421 | f.reset() |
| 422 | *logs = nil |
| 423 | |
| 424 | p.Set = nil |
| 425 | res, err := Converge(o, p) |
| 426 | if err != nil { |
| 427 | t.Fatalf("Converge (nil Set) returned an error: %v — leave-alone is not a failure", err) |
| 428 | } |
| 429 | if res.Changed { |
| 430 | t.Fatal("Changed = true on a nil Set") |
| 431 | } |
| 432 | if len(f.calls) != 0 { |
| 433 | t.Fatalf("systemctl calls on a nil Set: %v", f.calls) |
| 434 | } |
| 435 | if !bytes.Equal(readFile(t, svcPath), svcBefore) || !bytes.Equal(readFile(t, sockPath), sockBefore) { |
| 436 | t.Fatal("a nil Set changed the drop-ins") |
| 437 | } |
| 438 | if len(*logs) != 1 || !strings.Contains((*logs)[0], "leaving its drop-ins exactly as they are") { |
| 439 | t.Fatalf("logs = %v, want exactly one leave-alone line", *logs) |
| 440 | } |
| 441 | } |
| 442 | |
| 443 | // TestInvalidStagedManifestRefusesTheRouteAndLeavesDropInsAlone: a |
| 444 | // platform-binary rollback can meet an already-staged version whose |
| 445 | // capability block this binary does not understand (ensureStaged short- |
| 446 | // circuits on a staged artifact and never re-reads its manifest). That falls |
| 447 | // the same way as an unreadable manifest — nothing removed — but it also |
| 448 | // returns an error, so a version-change caller refuses the version rather |
| 449 | // than pointing live at something it cannot materialize. |
| 450 | func TestInvalidStagedManifestRefusesTheRouteAndLeavesDropInsAlone(t *testing.T) { |
| 451 | o, f, logs, users := testOps(t) |
| 452 | p := keymailPlan(t, keymailGrants()) |
| 453 | if _, err := Converge(o, p); err != nil { |
| 454 | t.Fatalf("Converge (valid): %v", err) |
| 455 | } |
| 456 | svcPath, sockPath := DropInPaths(o.EtcDir, p.Host) |
| 457 | svcBefore, sockBefore := readFile(t, svcPath), readFile(t, sockPath) |
| 458 | f.reset() |
| 459 | *logs = nil |
| 460 | |
| 461 | // A Set from a manifest a NEWER binary wrote: v2 is outside 1..Version. |
| 462 | // It declares host_role, which is granted — so if the re-validation is |
| 463 | // not the FIRST thing after the leave-alone branch, this refusal happens |
| 464 | // only at render time, after the pass has already minted a system account |
| 465 | // for a version it cannot materialize. |
| 466 | p.Set = &Set{V: Version + 1, Command: "relay", Daemon: true, HostRole: true} |
| 467 | res, err := Converge(o, p) |
| 468 | if err == nil { |
| 469 | t.Fatal("Converge accepted a capability block this binary does not speak") |
| 470 | } |
| 471 | if res.Changed { |
| 472 | t.Fatal("Changed = true on a refused route") |
| 473 | } |
| 474 | if len(f.calls) != 0 { |
| 475 | t.Fatalf("systemctl calls on a refused route: %v", f.calls) |
| 476 | } |
| 477 | if !bytes.Equal(readFile(t, svcPath), svcBefore) || !bytes.Equal(readFile(t, sockPath), sockBefore) { |
| 478 | t.Fatal("a refused route's drop-ins were modified") |
| 479 | } |
| 480 | if len(*users) != 1 { // the first, valid converge — and nothing since |
| 481 | t.Fatalf("EnsureUser calls = %v, want only the one from the valid pass", *users) |
| 482 | } |
| 483 | if !hasLog(*logs, "update the platform binary") { |
| 484 | t.Fatalf("logs = %v, want the update-the-platform-binary line", *logs) |
| 485 | } |
| 486 | } |
| 487 | |
| 488 | // TestHandWrittenDropInIsNeverOverwritten: a file at a generated path without |
| 489 | // the marker is never overwritten and never deleted — reported and skipped — |
| 490 | // while the OTHER unit's half still converges. |
| 491 | func TestHandWrittenDropInIsNeverOverwritten(t *testing.T) { |
| 492 | o, _, logs, _ := testOps(t) |
| 493 | p := keymailPlan(t, keymailGrants()) |
| 494 | svcPath, sockPath := DropInPaths(o.EtcDir, p.Host) |
| 495 | hand := []byte("[Service]\nEnvironment=HAND=written\n") |
| 496 | if err := os.MkdirAll(filepath.Dir(svcPath), 0o755); err != nil { |
| 497 | t.Fatal(err) |
| 498 | } |
| 499 | if err := os.WriteFile(svcPath, hand, 0o644); err != nil { |
| 500 | t.Fatal(err) |
| 501 | } |
| 502 | |
| 503 | if _, err := Converge(o, p); err != nil { |
| 504 | t.Fatalf("Converge: %v", err) |
| 505 | } |
| 506 | if got := readFile(t, svcPath); !bytes.Equal(got, hand) { |
| 507 | t.Fatalf("hand-written drop-in was modified:\n got: %q\nwant: %q", got, hand) |
| 508 | } |
| 509 | if !hasLog(*logs, "never overwritten, never removed") { |
| 510 | t.Fatalf("logs = %v, want the hand-edited report", *logs) |
| 511 | } |
| 512 | if got, want := readFile(t, sockPath), readGolden(t, "keymail.socket.conf"); !bytes.Equal(got, want) { |
| 513 | t.Fatalf("the other unit's half did not converge:\n got: %q\nwant: %q", got, want) |
| 514 | } |
| 515 | |
| 516 | // And it is not restorable either: this pass never wrote it, so it is not |
| 517 | // this pass's to put back — or to delete on a rollback. |
| 518 | res, _ := Converge(o, p) |
| 519 | if res.Prev.ServiceHad { |
| 520 | t.Fatal("Snapshot claimed a hand-written file as its own") |
| 521 | } |
| 522 | if err := Restore(o, res.Prev); err != nil { |
| 523 | t.Fatalf("Restore: %v", err) |
| 524 | } |
| 525 | if got := readFile(t, svcPath); !bytes.Equal(got, hand) { |
| 526 | t.Fatal("Restore deleted a hand-written drop-in") |
| 527 | } |
| 528 | } |
| 529 | |
| 530 | // TestUnchangedPassWritesNothing is the determinism promise, asserted against |
| 531 | // a DAEMON route with a socket — the case with the most unit state to |
| 532 | // repair. Once the box is converged, the hourly pass changes nothing at all. |
| 533 | func TestUnchangedPassWritesNothing(t *testing.T) { |
| 534 | o, f, logs, _ := testOps(t) |
| 535 | p := keymailPlan(t, keymailGrants()) |
| 536 | if _, err := Converge(o, p); err != nil { |
| 537 | t.Fatalf("Converge (first): %v", err) |
| 538 | } |
| 539 | svcPath, sockPath := DropInPaths(o.EtcDir, p.Host) |
| 540 | svcInfo, err := os.Stat(svcPath) |
| 541 | if err != nil { |
| 542 | t.Fatal(err) |
| 543 | } |
| 544 | f.reset() |
| 545 | *logs = nil |
| 546 | |
| 547 | res, err := Converge(o, p) |
| 548 | if err != nil { |
| 549 | t.Fatalf("Converge (second): %v", err) |
| 550 | } |
| 551 | if res.Changed { |
| 552 | t.Fatal("Changed = true on an unchanged pass") |
| 553 | } |
| 554 | if len(f.calls) != 0 { |
| 555 | t.Fatalf("an unchanged pass made systemctl calls: %v", f.calls) |
| 556 | } |
| 557 | if len(*logs) != 0 { |
| 558 | t.Fatalf("an unchanged pass logged: %v", *logs) |
| 559 | } |
| 560 | // Not even a rewrite with identical bytes: the mtime is untouched. |
| 561 | after, err := os.Stat(svcPath) |
| 562 | if err != nil { |
| 563 | t.Fatal(err) |
| 564 | } |
| 565 | if !after.ModTime().Equal(svcInfo.ModTime()) { |
| 566 | t.Fatal("the service drop-in was rewritten with identical bytes") |
| 567 | } |
| 568 | if got, want := readFile(t, sockPath), readGolden(t, "keymail.socket.conf"); !bytes.Equal(got, want) { |
| 569 | t.Fatalf("socket drop-in drifted on a no-op pass:\n got: %q\nwant: %q", got, want) |
| 570 | } |
| 571 | // The mode is part of the contract systemd reads. |
| 572 | if perm := svcInfo.Mode().Perm(); perm != 0o644 { |
| 573 | t.Fatalf("service drop-in mode = %o, want 0644", perm) |
| 574 | } |
| 575 | } |
| 576 | |
| 577 | // TestDaemonPlusHibernateIsRefused: a daemon the platform is asked to keep |
| 578 | // running and to stop when idle is a contradiction, and for these apps |
| 579 | // hibernation means an MX that sleeps. Refused authoritatively here, with the |
| 580 | // operator's exact fix in the line. |
| 581 | func TestDaemonPlusHibernateIsRefused(t *testing.T) { |
| 582 | o, f, logs, users := testOps(t) |
| 583 | p := keymailPlan(t, keymailGrants()) |
| 584 | p.Hibernate = true |
| 585 | |
| 586 | res, err := Converge(o, p) |
| 587 | if err == nil { |
| 588 | t.Fatal("Converge accepted daemon + hibernate") |
| 589 | } |
| 590 | if res.Changed { |
| 591 | t.Fatal("Changed = true on a refusal") |
| 592 | } |
| 593 | if len(f.calls) != 0 { |
| 594 | t.Fatalf("systemctl calls on a refusal: %v", f.calls) |
| 595 | } |
| 596 | if len(*users) != 0 { |
| 597 | t.Fatalf("a refused plan created users: %v", *users) |
| 598 | } |
| 599 | svcPath, sockPath := DropInPaths(o.EtcDir, p.Host) |
| 600 | mustNotExist(t, svcPath) |
| 601 | mustNotExist(t, sockPath) |
| 602 | if !hasLog(*logs, "carlos route --host keymail.dev --hibernate=false") { |
| 603 | t.Fatalf("logs = %v, want the --hibernate=false fix", *logs) |
| 604 | } |
| 605 | } |
| 606 | |
| 607 | // TestHostRoleFlipRefusedWhenTheStateDirIsTheOtherShape is spec §4b's pinned |
| 608 | // sharp edge: with DynamicUser=yes the state dir is a SYMLINK into |
| 609 | // /var/lib/private, and a static User= needs a real directory. The platform |
| 610 | // refuses in place; the operator does the stop → move → start. |
| 611 | func TestHostRoleFlipRefusedWhenTheStateDirIsTheOtherShape(t *testing.T) { |
| 612 | o, f, logs, users := testOps(t) |
| 613 | p := keymailPlan(t, keymailGrants()) |
| 614 | if err := os.Symlink(filepath.Join(o.StateRoot, "private", "carlos-app", p.Host), filepath.Join(o.StateRoot, p.Host)); err != nil { |
| 615 | t.Fatal(err) |
| 616 | } |
| 617 | |
| 618 | res, err := Converge(o, p) |
| 619 | if err == nil { |
| 620 | t.Fatal("Converge flipped host_role on over a DynamicUser state dir") |
| 621 | } |
| 622 | if res.Changed { |
| 623 | t.Fatal("Changed = true on a refusal") |
| 624 | } |
| 625 | if len(f.calls) != 0 { |
| 626 | t.Fatalf("systemctl calls on a refusal: %v", f.calls) |
| 627 | } |
| 628 | if len(*users) != 0 { |
| 629 | t.Fatalf("a refused plan created users: %v", *users) |
| 630 | } |
| 631 | svcPath, _ := DropInPaths(o.EtcDir, p.Host) |
| 632 | mustNotExist(t, svcPath) |
| 633 | if !hasLog(*logs, "refused in place") { |
| 634 | t.Fatalf("logs = %v, want a refused-in-place line naming the operator's move", *logs) |
| 635 | } |
| 636 | } |
| 637 | |
| 638 | // The OFF direction of the same flip: once THIS package's drop-in has put the |
| 639 | // unit on a static user, dropping host_role would hand the template's |
| 640 | // DynamicUser=yes an existing real directory it cannot use. |
| 641 | func TestHostRoleFlipOffRefusedWhenTheStateDirIsAlreadyStatic(t *testing.T) { |
| 642 | o, _, logs, _ := testOps(t) |
| 643 | p := keymailPlan(t, keymailGrants()) |
| 644 | if _, err := Converge(o, p); err != nil { |
| 645 | t.Fatalf("Converge (host_role on): %v", err) |
| 646 | } |
| 647 | if err := os.Mkdir(filepath.Join(o.StateRoot, p.Host), 0o700); err != nil { |
| 648 | t.Fatal(err) |
| 649 | } |
| 650 | svcPath, _ := DropInPaths(o.EtcDir, p.Host) |
| 651 | before := readFile(t, svcPath) |
| 652 | *logs = nil |
| 653 | |
| 654 | p.Grants = Grants{Ports: map[int]bool{25: true}} // host-role revoked |
| 655 | if _, err := Converge(o, p); err == nil { |
| 656 | t.Fatal("Converge flipped host_role off over a static state dir") |
| 657 | } |
| 658 | if got := readFile(t, svcPath); !bytes.Equal(got, before) { |
| 659 | t.Fatal("a refused flip rewrote the service drop-in") |
| 660 | } |
| 661 | if !hasLog(*logs, "refused in place") { |
| 662 | t.Fatalf("logs = %v, want a refused-in-place line", *logs) |
| 663 | } |
| 664 | } |
| 665 | |
| 666 | // A real state directory that no drop-in of OURS created is not a flip — it |
| 667 | // is a legacy hand-provisioned app — and must not freeze an unrelated route's |
| 668 | // capabilities forever. |
| 669 | func TestPreExistingStateDirDoesNotBlockARouteThatNeverHadHostRole(t *testing.T) { |
| 670 | o, _, _, _ := testOps(t) |
| 671 | var s Set |
| 672 | if err := s.UnmarshalJSON([]byte(`{"v":1,"command":"relay","daemon":true}`)); err != nil { |
| 673 | t.Fatal(err) |
| 674 | } |
| 675 | p := Plan{Host: "legacy.example.com", App: "legacy", Set: &s} |
| 676 | if err := os.Mkdir(filepath.Join(o.StateRoot, p.Host), 0o700); err != nil { |
| 677 | t.Fatal(err) |
| 678 | } |
| 679 | if _, err := Converge(o, p); err != nil { |
| 680 | t.Fatalf("Converge refused a route over a state dir it never created: %v", err) |
| 681 | } |
| 682 | svcPath, _ := DropInPaths(o.EtcDir, p.Host) |
| 683 | if !strings.Contains(string(readFile(t, svcPath)), "ExecStart=/opt/carlos/live/%i relay") { |
| 684 | t.Fatal("the route did not converge") |
| 685 | } |
| 686 | } |
| 687 | |
| 688 | // TestRestorePutsThePreviousDropInsBack is the drop-in half of adoptHost's |
| 689 | // restart-failure rollback (the symlink half is release.PointLive's). |
| 690 | func TestRestorePutsThePreviousDropInsBack(t *testing.T) { |
| 691 | o, f, _, _ := testOps(t) |
| 692 | p := keymailPlan(t, keymailGrants()) |
| 693 | if _, err := Converge(o, p); err != nil { |
| 694 | t.Fatalf("Converge A: %v", err) |
| 695 | } |
| 696 | svcPath, sockPath := DropInPaths(o.EtcDir, p.Host) |
| 697 | svcA, sockA := readFile(t, svcPath), readFile(t, sockPath) |
| 698 | |
| 699 | // Version B: the app dropped `ports` and `children` from its manifest. |
| 700 | var b Set |
| 701 | if err := b.UnmarshalJSON([]byte(`{"v":1,"command":"relay","daemon":true,"runtime_dir":"keymail","host_role":true}`)); err != nil { |
| 702 | t.Fatal(err) |
| 703 | } |
| 704 | p.Set = &b |
| 705 | resB, err := Converge(o, p) |
| 706 | if err != nil { |
| 707 | t.Fatalf("Converge B: %v", err) |
| 708 | } |
| 709 | if !resB.Changed { |
| 710 | t.Fatal("B did not change anything") |
| 711 | } |
| 712 | mustNotExist(t, sockPath) |
| 713 | f.reset() |
| 714 | |
| 715 | if err := Restore(o, resB.Prev); err != nil { |
| 716 | t.Fatalf("Restore: %v", err) |
| 717 | } |
| 718 | if got := readFile(t, svcPath); !bytes.Equal(got, svcA) { |
| 719 | t.Fatalf("service drop-in after Restore:\n got: %q\n want: %q", got, svcA) |
| 720 | } |
| 721 | if got := readFile(t, sockPath); !bytes.Equal(got, sockA) { |
| 722 | t.Fatalf("socket drop-in after Restore:\n got: %q\n want: %q", got, sockA) |
| 723 | } |
| 724 | reload := indexOf(f.calls, "daemon-reload") |
| 725 | if reload < 0 { |
| 726 | t.Fatalf("Restore did not daemon-reload (calls: %v)", f.calls) |
| 727 | } |
| 728 | // Bytes are not a binding. A restored ListenStream systemd never re-binds |
| 729 | // restores nothing — the production MX would stay off :25 after a rollback |
| 730 | // — so the socket half goes back through the pinned sequence, after the |
| 731 | // reload. |
| 732 | stop := indexOf(f.calls, "stop carlos-app@keymail.dev.service") |
| 733 | sock := indexOf(f.calls, "restart carlos-app@keymail.dev.socket") |
| 734 | start := indexOf(f.calls, "start carlos-app@keymail.dev.service") |
| 735 | if stop < 0 || sock < 0 || start < 0 || !(reload < stop && stop < sock && sock < start) { |
| 736 | t.Fatalf("Restore did not re-bind the restored listener in order: %v", f.calls) |
| 737 | } |
| 738 | } |
| 739 | |
| 740 | // The mirror case, and the one a bytes-only Restore gets exactly as wrong: |
| 741 | // version A had NO socket drop-in, B added one and bound :25, the restart |
| 742 | // failed, and the rollback must both delete the file and give the port back. |
| 743 | // Left bound, it is a public listener held by a binary that never asked for |
| 744 | // it and a claim that blocks the next route. |
| 745 | func TestRestoreReleasesAPortThatTheRolledBackVersionAdded(t *testing.T) { |
| 746 | o, f, _, _ := testOps(t) |
| 747 | var a Set |
| 748 | if err := a.UnmarshalJSON([]byte(`{"v":1,"command":"relay","daemon":true}`)); err != nil { |
| 749 | t.Fatal(err) |
| 750 | } |
| 751 | p := Plan{Host: "keymail.dev", App: "keymail", Set: &a, Grants: keymailGrants()} |
| 752 | if _, err := Converge(o, p); err != nil { |
| 753 | t.Fatalf("Converge A: %v", err) |
| 754 | } |
| 755 | _, sockPath := DropInPaths(o.EtcDir, p.Host) |
| 756 | mustNotExist(t, sockPath) |
| 757 | |
| 758 | p.Set = keymailSet(t) // B declares ports, and :25 is granted |
| 759 | resB, err := Converge(o, p) |
| 760 | if err != nil { |
| 761 | t.Fatalf("Converge B: %v", err) |
| 762 | } |
| 763 | if resB.Prev.SocketHad { |
| 764 | t.Fatal("the snapshot claims a socket drop-in that never existed") |
| 765 | } |
| 766 | readFile(t, sockPath) // B really did bind it |
| 767 | f.reset() |
| 768 | // The state a rollback actually happens in: the restart that triggered it |
| 769 | // failed, so the service is DOWN. Whether it is supposed to be running is |
| 770 | // not something the disk can answer, which is why the snapshot carries it |
| 771 | // — without that, the sequence would skip the stop/start pair and leave a |
| 772 | // daemon stopped on the version it was just rolled back to. |
| 773 | f.active[ServiceUnit(p.Host)] = false |
| 774 | |
| 775 | if err := Restore(o, resB.Prev); err != nil { |
| 776 | t.Fatalf("Restore: %v", err) |
| 777 | } |
| 778 | mustNotExist(t, sockPath) |
| 779 | reload := indexOf(f.calls, "daemon-reload") |
| 780 | sock := indexOf(f.calls, "restart carlos-app@keymail.dev.socket") |
| 781 | if reload < 0 || sock < 0 || reload > sock { |
| 782 | t.Fatalf("Restore removed the drop-in but never released the port: %v", f.calls) |
| 783 | } |
| 784 | if indexOf(f.calls, "start carlos-app@keymail.dev.service") < 0 || !f.active[ServiceUnit(p.Host)] { |
| 785 | t.Fatalf("Restore left the daemon stopped: %v", f.calls) |
| 786 | } |
| 787 | // And the box no longer counts it as a claim, so the next route can have it. |
| 788 | if ports, _ := scanClaims(o.EtcDir, "other.example.com"); ports[25] != "" { |
| 789 | t.Fatalf("port 25 is still claimed by %q after the rollback", ports[25]) |
| 790 | } |
| 791 | } |
| 792 | |
| 793 | // TestCheckClaimsRefusesAPortHeldByTwoRoutesAndARuntimeDirCollision: box-wide |
| 794 | // exclusivity, resolved before anything is written, and deterministic on ties |
| 795 | // so a conflict does not flap between passes. |
| 796 | func TestCheckClaimsRefusesAPortHeldByTwoRoutesAndARuntimeDirCollision(t *testing.T) { |
| 797 | mx := func(host string) Plan { |
| 798 | var s Set |
| 799 | if err := s.UnmarshalJSON([]byte(`{"v":1,"ports":[{"port":25,"name":"mx"}]}`)); err != nil { |
| 800 | t.Fatal(err) |
| 801 | } |
| 802 | return Plan{Host: host, App: "keymail", Set: &s, Grants: Grants{Ports: map[int]bool{25: true}}} |
| 803 | } |
| 804 | rd := func(host string) Plan { |
| 805 | var s Set |
| 806 | if err := s.UnmarshalJSON([]byte(`{"v":1,"runtime_dir":"keymail"}`)); err != nil { |
| 807 | t.Fatal(err) |
| 808 | } |
| 809 | return Plan{Host: host, App: "keymail", Set: &s} |
| 810 | } |
| 811 | |
| 812 | for i := 0; i < 5; i++ { // deterministic across repeated runs, both input orders |
| 813 | plans := []Plan{mx("b.example.com"), mx("a.example.com"), rd("d.example.com"), rd("c.example.com")} |
| 814 | if i%2 == 1 { |
| 815 | plans = []Plan{plans[1], plans[0], plans[3], plans[2]} |
| 816 | } |
| 817 | refused := CheckClaims(plans) |
| 818 | if len(refused) != 2 { |
| 819 | t.Fatalf("refused %d routes, want 2: %v", len(refused), refused) |
| 820 | } |
| 821 | if refused["a.example.com"] != nil || refused["c.example.com"] != nil { |
| 822 | t.Fatalf("the lexicographically-first host lost its claim: %v", refused) |
| 823 | } |
| 824 | if refused["b.example.com"] == nil || !strings.Contains(refused["b.example.com"].Error(), "a.example.com") { |
| 825 | t.Fatalf("port conflict not refused against the holder: %v", refused["b.example.com"]) |
| 826 | } |
| 827 | if refused["d.example.com"] == nil || !strings.Contains(refused["d.example.com"].Error(), "c.example.com") { |
| 828 | t.Fatalf("runtime_dir conflict not refused against the holder: %v", refused["d.example.com"]) |
| 829 | } |
| 830 | } |
| 831 | } |
| 832 | |
| 833 | // An UNGRANTED port claims nothing: nothing materializes for it, so it cannot |
| 834 | // lock another route out of the port it was actually granted. |
| 835 | func TestCheckClaimsIgnoresUngrantedPortsAndUnresolvedManifests(t *testing.T) { |
| 836 | var s Set |
| 837 | if err := s.UnmarshalJSON([]byte(`{"v":1,"ports":[{"port":25,"name":"mx"}]}`)); err != nil { |
| 838 | t.Fatal(err) |
| 839 | } |
| 840 | plans := []Plan{ |
| 841 | {Host: "a.example.com", App: "keymail", Set: &s}, // declared, not granted |
| 842 | {Host: "b.example.com", App: "keymail", Set: &s, Grants: Grants{Ports: map[int]bool{25: true}}}, // granted |
| 843 | {Host: "c.example.com", App: "keymail", Set: nil}, // leave-alone claims nothing |
| 844 | } |
| 845 | if refused := CheckClaims(plans); len(refused) != 0 { |
| 846 | t.Fatalf("refused %v, want none", refused) |
| 847 | } |
| 848 | } |
| 849 | |
| 850 | // TestPortClaimedByAnotherRoutesDropInIsRefused covers the single-route |
| 851 | // version-change path, which never sees the other plans: the claim is read |
| 852 | // off what is already written on the box, so a stale drop-in for a route that |
| 853 | // has left the registry still holds its port. |
| 854 | func TestPortClaimedByAnotherRoutesDropInIsRefused(t *testing.T) { |
| 855 | o, f, logs, _ := testOps(t) |
| 856 | first := keymailPlan(t, keymailGrants()) |
| 857 | if _, err := Converge(o, first); err != nil { |
| 858 | t.Fatalf("Converge (first): %v", err) |
| 859 | } |
| 860 | f.reset() |
| 861 | *logs = nil |
| 862 | |
| 863 | // A second MX-shaped route with NO runtime_dir, so the port claim is the |
| 864 | // only thing that can refuse it. |
| 865 | var s Set |
| 866 | if err := s.UnmarshalJSON([]byte(`{"v":1,"command":"relay","daemon":true,"ports":[{"port":25,"name":"mx"}]}`)); err != nil { |
| 867 | t.Fatal(err) |
| 868 | } |
| 869 | second := Plan{Host: "mx2.example.com", App: "keymail", Set: &s, Grants: Grants{Ports: map[int]bool{25: true}}} |
| 870 | res, err := Converge(o, second) |
| 871 | if err == nil { |
| 872 | t.Fatal("two routes were allowed to claim :25 on one box") |
| 873 | } |
| 874 | if res.Changed { |
| 875 | t.Fatal("Changed = true on a refusal") |
| 876 | } |
| 877 | if len(f.calls) != 0 { |
| 878 | t.Fatalf("systemctl calls on a refusal: %v", f.calls) |
| 879 | } |
| 880 | svcPath, sockPath := DropInPaths(o.EtcDir, second.Host) |
| 881 | mustNotExist(t, svcPath) |
| 882 | mustNotExist(t, sockPath) |
| 883 | if !hasLog(*logs, "port 25 is already claimed by keymail.dev") { |
| 884 | t.Fatalf("logs = %v, want the port-conflict line naming the port and the holder", *logs) |
| 885 | } |
| 886 | |
| 887 | // The same route reconverging keeps its own claim: it is not in conflict |
| 888 | // with itself. |
| 889 | if _, err := Converge(o, first); err != nil { |
| 890 | t.Fatalf("the holder refused its own claim: %v", err) |
| 891 | } |
| 892 | } |
| 893 | |
| 894 | // The same disk-read claim check for runtime_dir: the second app to claim |
| 895 | // `keymail` is refused rather than handed a directory another app's sockets |
| 896 | // live in (spec §7). |
| 897 | func TestRuntimeDirClaimedByAnotherRoutesDropInIsRefused(t *testing.T) { |
| 898 | o, f, logs, _ := testOps(t) |
| 899 | if _, err := Converge(o, keymailPlan(t, keymailGrants())); err != nil { |
| 900 | t.Fatalf("Converge (first): %v", err) |
| 901 | } |
| 902 | f.reset() |
| 903 | *logs = nil |
| 904 | |
| 905 | var s Set |
| 906 | if err := s.UnmarshalJSON([]byte(`{"v":1,"command":"serve","runtime_dir":"keymail"}`)); err != nil { |
| 907 | t.Fatal(err) |
| 908 | } |
| 909 | second := Plan{Host: "other.example.com", App: "other", Set: &s} |
| 910 | if _, err := Converge(o, second); err == nil { |
| 911 | t.Fatal("two routes were allowed to claim /run/keymail on one box") |
| 912 | } |
| 913 | if len(f.calls) != 0 { |
| 914 | t.Fatalf("systemctl calls on a refusal: %v", f.calls) |
| 915 | } |
| 916 | svcPath, _ := DropInPaths(o.EtcDir, second.Host) |
| 917 | mustNotExist(t, svcPath) |
| 918 | if !hasLog(*logs, `runtime_dir "keymail" is already claimed by keymail.dev`) { |
| 919 | t.Fatalf("logs = %v, want the runtime_dir-conflict line naming the holder", *logs) |
| 920 | } |
| 921 | } |
| 922 | |
| 923 | // A hand-written drop-in at another host's generated path still holds its |
| 924 | // port: this pass will never rewrite it, so systemd applies it, so the claim |
| 925 | // is real. Fail closed — the alternative is two units racing for :25. |
| 926 | func TestAHandWrittenDropInStillHoldsItsPortAgainstAnotherRoute(t *testing.T) { |
| 927 | o, _, logs, _ := testOps(t) |
| 928 | svcPath, _ := DropInPaths(o.EtcDir, "hand.example.com") |
| 929 | if err := os.MkdirAll(filepath.Dir(svcPath), 0o755); err != nil { |
| 930 | t.Fatal(err) |
| 931 | } |
| 932 | sockPath := filepath.Join(o.EtcDir, SocketUnit("hand.example.com")+".d", DropInName) |
| 933 | if err := os.MkdirAll(filepath.Dir(sockPath), 0o755); err != nil { |
| 934 | t.Fatal(err) |
| 935 | } |
| 936 | if err := os.WriteFile(sockPath, []byte("[Socket]\nListenStream=25\n"), 0o644); err != nil { |
| 937 | t.Fatal(err) |
| 938 | } |
| 939 | |
| 940 | if _, err := Converge(o, keymailPlan(t, keymailGrants())); err == nil { |
| 941 | t.Fatal("a route was granted a port a hand-written drop-in already binds") |
| 942 | } |
| 943 | if !hasLog(*logs, "port 25 is already claimed by hand.example.com") { |
| 944 | t.Fatalf("logs = %v, want the port-conflict line", *logs) |
| 945 | } |
| 946 | } |
| 947 | |
| 948 | // Changed is true when only a REMOVAL happened: the caller's rollback and the |
| 949 | // pass's own daemon-reload both hang off it, and "the box changed" is as true |
| 950 | // of a deletion as of a write. |
| 951 | func TestChangedIsTrueWhenOnlyARemovalHappened(t *testing.T) { |
| 952 | o, f, _, _ := testOps(t) |
| 953 | p := keymailPlan(t, keymailGrants()) |
| 954 | if _, err := Converge(o, p); err != nil { |
| 955 | t.Fatalf("Converge (first): %v", err) |
| 956 | } |
| 957 | svcPath, sockPath := DropInPaths(o.EtcDir, p.Host) |
| 958 | f.reset() |
| 959 | |
| 960 | // A new version declaring only self-service, directive-free capabilities: |
| 961 | // both halves render nil, so both generated files go away. |
| 962 | var s Set |
| 963 | if err := s.UnmarshalJSON([]byte(`{"v":1,"daemon":true,"data":{"replicate":"platform","dbs":["registry.db"]}}`)); err != nil { |
| 964 | t.Fatal(err) |
| 965 | } |
| 966 | p.Set = &s |
| 967 | res, err := Converge(o, p) |
| 968 | if err != nil { |
| 969 | t.Fatalf("Converge (removal): %v", err) |
| 970 | } |
| 971 | if !res.Changed { |
| 972 | t.Fatal("Changed = false when both drop-ins were removed") |
| 973 | } |
| 974 | mustNotExist(t, svcPath) |
| 975 | mustNotExist(t, sockPath) |
| 976 | if indexOf(f.calls, "daemon-reload") < 0 { |
| 977 | t.Fatalf("a removal did not daemon-reload: %v", f.calls) |
| 978 | } |
| 979 | } |
| 980 | |
| 981 | // The /run existence check is the SECOND runtime_dir guard: the static |
| 982 | // reserved list is finite, and a box can have anything under /run that no |
| 983 | // list will ever enumerate. |
| 984 | func TestRuntimeDirRefusedWhenRunPathAlreadyExists(t *testing.T) { |
| 985 | o, f, logs, _ := testOps(t) |
| 986 | p := keymailPlan(t, keymailGrants()) |
| 987 | if err := os.Mkdir(filepath.Join(o.RunDir, "keymail"), 0o755); err != nil { |
| 988 | t.Fatal(err) |
| 989 | } |
| 990 | |
| 991 | if _, err := Converge(o, p); err == nil { |
| 992 | t.Fatal("Converge accepted a runtime_dir whose /run path already exists") |
| 993 | } |
| 994 | if len(f.calls) != 0 { |
| 995 | t.Fatalf("systemctl calls on a refusal: %v", f.calls) |
| 996 | } |
| 997 | svcPath, _ := DropInPaths(o.EtcDir, p.Host) |
| 998 | mustNotExist(t, svcPath) |
| 999 | if !hasLog(*logs, "is not this route's own claim") { |
| 1000 | t.Fatalf("logs = %v, want the /run existence refusal", *logs) |
| 1001 | } |
| 1002 | } |
| 1003 | |
| 1004 | // ...and it must not refuse a route its OWN previous pass created the |
| 1005 | // directory for, or a converged box would refuse itself every hour. |
| 1006 | func TestRuntimeDirAcceptsThisRoutesOwnPriorClaim(t *testing.T) { |
| 1007 | o, _, _, _ := testOps(t) |
| 1008 | p := keymailPlan(t, keymailGrants()) |
| 1009 | if _, err := Converge(o, p); err != nil { |
| 1010 | t.Fatalf("Converge (first): %v", err) |
| 1011 | } |
| 1012 | // systemd creates /run/keymail once the unit starts. |
| 1013 | if err := os.Mkdir(filepath.Join(o.RunDir, "keymail"), 0o755); err != nil { |
| 1014 | t.Fatal(err) |
| 1015 | } |
| 1016 | if _, err := Converge(o, p); err != nil { |
| 1017 | t.Fatalf("Converge refused the route its own RuntimeDirectory belongs to: %v", err) |
| 1018 | } |
| 1019 | } |
| 1020 | |
| 1021 | // TestHostWithASeparatorIsRefused: a registry row is trusted at the same |
| 1022 | // level as live/<host>, but this one becomes a filename under |
| 1023 | // /etc/systemd/system, written as root (appconfig.syncHost's guard, same |
| 1024 | // direction). |
| 1025 | func TestHostWithASeparatorIsRefused(t *testing.T) { |
| 1026 | o, f, _, _ := testOps(t) |
| 1027 | for _, host := range []string{"", ".", "..", "../../etc/passwd", "a/b", "sub/../keymail.dev"} { |
| 1028 | p := keymailPlan(t, keymailGrants()) |
| 1029 | p.Host = host |
| 1030 | if _, err := Converge(o, p); err == nil { |
| 1031 | t.Fatalf("Converge accepted host %q", host) |
| 1032 | } |
| 1033 | if err := Restore(o, Snapshot{Host: host}); err == nil { |
| 1034 | t.Fatalf("Restore accepted host %q", host) |
| 1035 | } |
| 1036 | if err := Restart(o, p); err == nil { |
| 1037 | t.Fatalf("Restart accepted host %q", host) |
| 1038 | } |
| 1039 | } |
| 1040 | if len(f.calls) != 0 { |
| 1041 | t.Fatalf("a refused host reached systemctl: %v", f.calls) |
| 1042 | } |
| 1043 | ents, err := os.ReadDir(o.EtcDir) |
| 1044 | if err != nil { |
| 1045 | t.Fatal(err) |
| 1046 | } |
| 1047 | if len(ents) != 0 { |
| 1048 | t.Fatalf("a refused host wrote into EtcDir: %v", ents) |
| 1049 | } |
| 1050 | } |
| 1051 | |
| 1052 | // TestSocketRestartFailureStillStartsTheService: the service must never be |
| 1053 | // left stopped because the socket failed to come back — and the error must |
| 1054 | // still be returned, so the caller's rollback fires. |
| 1055 | func TestSocketRestartFailureStillStartsTheService(t *testing.T) { |
| 1056 | o, f, _, _ := testOps(t) |
| 1057 | p := keymailPlan(t, keymailGrants()) |
| 1058 | f.fail["restart carlos-app@keymail.dev.socket"] = errors.New("Address already in use") |
| 1059 | |
| 1060 | res, err := Converge(o, p) |
| 1061 | if err == nil { |
| 1062 | t.Fatal("Converge swallowed a failed socket restart") |
| 1063 | } |
| 1064 | if !strings.Contains(err.Error(), "Address already in use") { |
| 1065 | t.Fatalf("error = %v, want the socket failure", err) |
| 1066 | } |
| 1067 | // systemd's listener no longer matches the bytes, so the bytes go back: |
| 1068 | // otherwise the next pass finds no diff and never retries, and the box |
| 1069 | // sits on a drop-in that reads "converged" forever. |
| 1070 | if res.Changed { |
| 1071 | t.Fatal("Changed = true, but the failed pass was rolled back") |
| 1072 | } |
| 1073 | svcPath, sockPath := DropInPaths(o.EtcDir, p.Host) |
| 1074 | mustNotExist(t, svcPath) |
| 1075 | mustNotExist(t, sockPath) |
| 1076 | stop := indexOf(f.calls, "stop carlos-app@keymail.dev.service") |
| 1077 | start := indexOf(f.calls, "start carlos-app@keymail.dev.service") |
| 1078 | if stop < 0 || start < 0 || start < stop { |
| 1079 | t.Fatalf("the service was left stopped after a failed socket restart: %v", f.calls) |
| 1080 | } |
| 1081 | if !f.active[ServiceUnit(p.Host)] { |
| 1082 | t.Fatalf("the service is not running after the sequence: %v", f.calls) |
| 1083 | } |
| 1084 | |
| 1085 | // The same failure on a NON-daemon route, where no enable/start repair |
| 1086 | // exists to paper over it: the start can only have come from the sequence |
| 1087 | // itself. This is the assertion that pins "the start is attempted even |
| 1088 | // when the socket verb failed" rather than merely observing a start. |
| 1089 | o2, f2, _, _ := testOps(t) |
| 1090 | var s Set |
| 1091 | if err := s.UnmarshalJSON([]byte(`{"v":1,"command":"relay","ports":[{"port":25,"name":"mx"}]}`)); err != nil { |
| 1092 | t.Fatal(err) |
| 1093 | } |
| 1094 | q := Plan{Host: "mx.example.com", App: "keymail", Set: &s, Grants: Grants{Ports: map[int]bool{25: true}}} |
| 1095 | f2.active[ServiceUnit(q.Host)] = true // running, so the sequence stops it |
| 1096 | f2.fail["restart carlos-app@mx.example.com.socket"] = errors.New("Address already in use") |
| 1097 | |
| 1098 | if _, err := Converge(o2, q); err == nil { |
| 1099 | t.Fatal("Converge swallowed a failed socket restart on a non-daemon route") |
| 1100 | } |
| 1101 | stop = indexOf(f2.calls, "stop carlos-app@mx.example.com.service") |
| 1102 | start = indexOf(f2.calls, "start carlos-app@mx.example.com.service") |
| 1103 | if stop < 0 || start < 0 || start < stop { |
| 1104 | t.Fatalf("a non-daemon service was left stopped after a failed socket restart: %v", f2.calls) |
| 1105 | } |
| 1106 | if !f2.active[ServiceUnit(q.Host)] { |
| 1107 | t.Fatalf("the service is not running after the sequence: %v", f2.calls) |
| 1108 | } |
| 1109 | } |
| 1110 | |
| 1111 | // EnsureUser failing refuses the WHOLE plan: writing User=carlos-keymail for |
| 1112 | // a user that does not exist gives systemd a unit that fails to start, which |
| 1113 | // Restart=always then spins. |
| 1114 | func TestEnsureUserFailureRefusesThePlan(t *testing.T) { |
| 1115 | o, f, logs, _ := testOps(t) |
| 1116 | o.EnsureUser = func(string) error { return errors.New("useradd: no free uid") } |
| 1117 | p := keymailPlan(t, keymailGrants()) |
| 1118 | |
| 1119 | if _, err := Converge(o, p); err == nil { |
| 1120 | t.Fatal("Converge wrote a User= line for a user it could not create") |
| 1121 | } |
| 1122 | svcPath, sockPath := DropInPaths(o.EtcDir, p.Host) |
| 1123 | mustNotExist(t, svcPath) |
| 1124 | mustNotExist(t, sockPath) |
| 1125 | if len(f.calls) != 0 { |
| 1126 | t.Fatalf("systemctl calls after a user-creation failure: %v", f.calls) |
| 1127 | } |
| 1128 | if !hasLog(*logs, "refusing the whole plan") { |
| 1129 | t.Fatalf("logs = %v, want the refusal", *logs) |
| 1130 | } |
| 1131 | } |
| 1132 | |
| 1133 | // The user is created only when host_role is BOTH declared and granted — an |
| 1134 | // ungranted declaration must not mint a system account. |
| 1135 | func TestUngrantedHostRoleCreatesNoUser(t *testing.T) { |
| 1136 | o, _, _, users := testOps(t) |
| 1137 | p := keymailPlan(t, Grants{}) |
| 1138 | if _, err := Converge(o, p); err != nil { |
| 1139 | t.Fatalf("Converge: %v", err) |
| 1140 | } |
| 1141 | if len(*users) != 0 { |
| 1142 | t.Fatalf("EnsureUser called for an ungranted host_role: %v", *users) |
| 1143 | } |
| 1144 | } |
| 1145 | |
| 1146 | // A .d path that is a FILE rather than a directory is an honest error, not a |
| 1147 | // panic and not a silent skip. |
| 1148 | func TestDropInDirectoryThatIsAFileIsAnError(t *testing.T) { |
| 1149 | o, _, _, _ := testOps(t) |
| 1150 | p := keymailPlan(t, keymailGrants()) |
| 1151 | svcPath, _ := DropInPaths(o.EtcDir, p.Host) |
| 1152 | if err := os.WriteFile(filepath.Dir(svcPath), []byte("not a directory"), 0o644); err != nil { |
| 1153 | t.Fatal(err) |
| 1154 | } |
| 1155 | if _, err := Converge(o, p); err == nil { |
| 1156 | t.Fatal("Converge succeeded with a file where the .d directory belongs") |
| 1157 | } |
| 1158 | } |
| 1159 | |
| 1160 | // A non-daemon route's restart stays the caller's try-restart, byte- |
| 1161 | // identically to today: converge writes the unit text and starts nothing. |
| 1162 | func TestNonDaemonRouteIsNeitherEnabledNorStarted(t *testing.T) { |
| 1163 | o, f, _, _ := testOps(t) |
| 1164 | var s Set |
| 1165 | if err := s.UnmarshalJSON([]byte(`{"v":1,"command":"serve","children":true}`)); err != nil { |
| 1166 | t.Fatal(err) |
| 1167 | } |
| 1168 | p := Plan{Host: "app.example.com", App: "app", Set: &s, Hibernate: true} |
| 1169 | |
| 1170 | if _, err := Converge(o, p); err != nil { |
| 1171 | t.Fatalf("Converge: %v", err) |
| 1172 | } |
| 1173 | for _, c := range f.calls { |
| 1174 | if c != "daemon-reload" { |
| 1175 | t.Fatalf("a hibernating non-daemon route had its unit state changed: %v", f.calls) |
| 1176 | } |
| 1177 | } |
| 1178 | svcPath, _ := DropInPaths(o.EtcDir, p.Host) |
| 1179 | if !strings.Contains(string(readFile(t, svcPath)), "ExecStart=/opt/carlos/live/%i serve") { |
| 1180 | t.Fatal("the drop-in was not written for a hibernating route") |
| 1181 | } |
| 1182 | } |
| 1183 | |
| 1184 | // Drift repair, the other half of the determinism promise: a converged box |
| 1185 | // whose daemon has been stopped and whose units have been disabled by hand is |
| 1186 | // put back — this is what the hourly pass is for (spec §4d). |
| 1187 | func TestStoppedDaemonAndDisabledUnitsAreRepaired(t *testing.T) { |
| 1188 | o, f, _, _ := testOps(t) |
| 1189 | p := keymailPlan(t, keymailGrants()) |
| 1190 | if _, err := Converge(o, p); err != nil { |
| 1191 | t.Fatalf("Converge (first): %v", err) |
| 1192 | } |
| 1193 | f.active[ServiceUnit(p.Host)] = false |
| 1194 | f.enabled[ServiceUnit(p.Host)] = false |
| 1195 | f.enabled[SocketUnit(p.Host)] = false |
| 1196 | f.reset() |
| 1197 | |
| 1198 | res, err := Converge(o, p) |
| 1199 | if err != nil { |
| 1200 | t.Fatalf("Converge (repair): %v", err) |
| 1201 | } |
| 1202 | if res.Changed { |
| 1203 | t.Fatal("Changed = true, but no byte moved — Changed is about the FILES") |
| 1204 | } |
| 1205 | if indexOf(f.calls, "daemon-reload") >= 0 { |
| 1206 | t.Fatalf("a unit-state repair reloaded the daemon: %v", f.calls) |
| 1207 | } |
| 1208 | for _, want := range []string{ |
| 1209 | "enable carlos-app@keymail.dev.socket", |
| 1210 | "enable carlos-app@keymail.dev.service", |
| 1211 | "start carlos-app@keymail.dev.service", |
| 1212 | } { |
| 1213 | if indexOf(f.calls, want) < 0 { |
| 1214 | t.Fatalf("drift not repaired: %q missing from %v", want, f.calls) |
| 1215 | } |
| 1216 | } |
| 1217 | } |
| 1218 | |
| 1219 | // Restart's error is REAL — it is what makes adoptHost's rollback fire. |
| 1220 | func TestRestartReportsItsFailure(t *testing.T) { |
| 1221 | o, f, _, _ := testOps(t) |
| 1222 | p := keymailPlan(t, keymailGrants()) |
| 1223 | f.fail["restart carlos-app@keymail.dev.service"] = errors.New("job failed") |
| 1224 | if err := Restart(o, p); err == nil { |
| 1225 | t.Fatal("Restart swallowed a failed restart") |
| 1226 | } |
| 1227 | if f.calls[len(f.calls)-1] != "restart carlos-app@keymail.dev.service" { |
| 1228 | t.Fatalf("Restart ran %v", f.calls) |
| 1229 | } |
| 1230 | } |
| 1231 | |
| 1232 | // Two plans, one box, no conflict: the second route's drop-ins are written |
| 1233 | // and the first route's are untouched. |
| 1234 | func TestASecondRouteOnTheBoxConvergesIndependently(t *testing.T) { |
| 1235 | o, _, _, _ := testOps(t) |
| 1236 | first := keymailPlan(t, keymailGrants()) |
| 1237 | if _, err := Converge(o, first); err != nil { |
| 1238 | t.Fatalf("Converge (first): %v", err) |
| 1239 | } |
| 1240 | svcA, sockA := DropInPaths(o.EtcDir, first.Host) |
| 1241 | beforeSvc, beforeSock := readFile(t, svcA), readFile(t, sockA) |
| 1242 | |
| 1243 | second := Plan{Host: "platform.carlosframework.com", App: "console", Set: consoleSet(t), Grants: consoleGrants()} |
| 1244 | if _, err := Converge(o, second); err != nil { |
| 1245 | t.Fatalf("Converge (second): %v", err) |
| 1246 | } |
| 1247 | svcB, sockB := DropInPaths(o.EtcDir, second.Host) |
| 1248 | if got, want := readFile(t, svcB), readGolden(t, "console.service.conf"); !bytes.Equal(got, want) { |
| 1249 | t.Fatalf("console service drop-in:\n got: %q\n want: %q", got, want) |
| 1250 | } |
| 1251 | mustNotExist(t, sockB) // console declares no ports |
| 1252 | if !bytes.Equal(readFile(t, svcA), beforeSvc) || !bytes.Equal(readFile(t, sockA), beforeSock) { |
| 1253 | t.Fatal("converging one route changed another's drop-ins") |
| 1254 | } |
| 1255 | } |
| 1256 | |
amadan