| 1 | //go:build linux |
| 2 | |
| 3 | package host |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "errors" |
| 8 | "net" |
| 9 | "os" |
| 10 | "path/filepath" |
| 11 | "reflect" |
| 12 | "testing" |
| 13 | |
| 14 | "github.com/carlosframework/platform/internal/hostbox" |
| 15 | "github.com/carlosframework/platform/internal/provision" |
| 16 | "golang.org/x/sys/unix" |
| 17 | ) |
| 18 | |
| 19 | func TestServingNamespaceAuthenticatedParentAndRecovery(t *testing.T) { |
| 20 | f := newServingRootFixture(t, true) |
| 21 | fs := f.p.ledger.fs.(*servingParkFS) |
| 22 | path := filepath.Join(fs.data, "instances", ".serving-parked") |
| 23 | if err := os.Mkdir(path, 0700); err != nil { |
| 24 | t.Fatal(err) |
| 25 | } |
| 26 | if err := os.WriteFile(filepath.Join(path, "unknown"), []byte("keep"), 0600); err != nil { |
| 27 | t.Fatal(err) |
| 28 | } |
| 29 | before := f.f.read() |
| 30 | a, err := f.park() |
| 31 | if err != nil { |
| 32 | t.Fatal(err) |
| 33 | } |
| 34 | anchor, err := fs.namespaceParent() |
| 35 | if err != nil { |
| 36 | t.Fatal(err) |
| 37 | } |
| 38 | r, err := f.p.ledger.read(a.ID) |
| 39 | if err != nil || anchor.Parent != r.Source.Parent { |
| 40 | t.Fatal("parent was not captured from the root installation", err) |
| 41 | } |
| 42 | var q servingParkQuarantine |
| 43 | body, err := f.disk.Read("namespace-quarantine.json") |
| 44 | if err != nil || servingParkDecode(body, &q) != nil || !q.Complete { |
| 45 | t.Fatal("authenticated recovery", err) |
| 46 | } |
| 47 | body, err = os.ReadFile(filepath.Join(filepath.Dir(path), q.name(), "unknown")) |
| 48 | if err != nil || string(body) != "keep" { |
| 49 | t.Fatal("unknown bytes changed", err) |
| 50 | } |
| 51 | if !reflect.DeepEqual(f.f.read(), before) { |
| 52 | t.Fatal("namespace recovery changed the claim") |
| 53 | } |
| 54 | } |
| 55 | |
| 56 | func TestServingNamespaceParentRequiresIndependentInstallation(t *testing.T) { |
| 57 | for _, fault := range []string{"peer", "census", "installation", "inventory", "receipt", "source", "mode"} { |
| 58 | t.Run(fault, func(t *testing.T) { |
| 59 | f := newServingRootFixture(t, true) |
| 60 | fs := f.p.ledger.fs.(*servingParkFS) |
| 61 | path := filepath.Join(fs.data, "instances", ".serving-parked") |
| 62 | if err := os.Mkdir(path, 0700); err != nil { |
| 63 | t.Fatal(err) |
| 64 | } |
| 65 | switch fault { |
| 66 | case "peer": |
| 67 | f.p.root.Stop.Authenticate = func(context.Context, net.Conn) (StopPeer, error) { return StopPeer{}, hostbox.ErrChanged } |
| 68 | case "census": |
| 69 | f.p.root.RequireNewWork = func(context.Context) error { return hostbox.ErrChanged } |
| 70 | case "installation": |
| 71 | f.actor.Enrollment = "changed" |
| 72 | case "inventory": |
| 73 | f.actor.Binding = "8" |
| 74 | case "receipt": |
| 75 | f.mutateRoot(func(m map[string]any) { m["State"].(map[string]any)["fence_journal"] = []any{} }) |
| 76 | case "source": |
| 77 | if err := os.Rename(filepath.Dir(f.db), filepath.Dir(f.db)+"-saved"); err != nil { |
| 78 | t.Fatal(err) |
| 79 | } |
| 80 | if err := os.Mkdir(filepath.Dir(f.db), 0700); err != nil { |
| 81 | t.Fatal(err) |
| 82 | } |
| 83 | case "mode": |
| 84 | if err := os.Chmod(filepath.Dir(path), 0777); err != nil { |
| 85 | t.Fatal(err) |
| 86 | } |
| 87 | } |
| 88 | if _, err := f.park(); err == nil { |
| 89 | t.Fatal("unverified recovery accepted") |
| 90 | } |
| 91 | for _, name := range []string{"namespace-parent.json", "namespace.json", "namespace-quarantine.json"} { |
| 92 | if _, err := f.disk.Read(name); !errors.Is(err, os.ErrNotExist) { |
| 93 | t.Fatal("unverified recovery wrote", name, err) |
| 94 | } |
| 95 | } |
| 96 | if _, err := os.Stat(path); err != nil { |
| 97 | t.Fatal("obstruction moved without authority", err) |
| 98 | } |
| 99 | }) |
| 100 | } |
| 101 | } |
| 102 | |
| 103 | func TestServingNamespaceStartupUsesRecordedParent(t *testing.T) { |
| 104 | for _, boundary := range []string{"parent", "quarantine-rename", "fresh-before", "fresh", "complete"} { |
| 105 | t.Run(boundary, func(t *testing.T) { |
| 106 | f := newServingRootFixture(t, true) |
| 107 | fs := f.p.ledger.fs.(*servingParkFS) |
| 108 | path := filepath.Join(fs.data, "instances", ".serving-parked") |
| 109 | if err := os.Mkdir(path, 0700); err != nil { |
| 110 | t.Fatal(err) |
| 111 | } |
| 112 | if err := os.WriteFile(filepath.Join(path, "unknown"), []byte("retained"), 0600); err != nil { |
| 113 | t.Fatal(err) |
| 114 | } |
| 115 | fs.state = quarantineWriteFault{Store: f.disk, before: func(name string, body []byte) error { |
| 116 | var q servingParkQuarantine |
| 117 | if boundary == "fresh-before" && name == "namespace-quarantine.json" && servingParkDecode(body, &q) == nil && q.Fresh.Inode != 0 { |
| 118 | return unix.EIO |
| 119 | } |
| 120 | return nil |
| 121 | }, after: func(name string, body []byte) error { |
| 122 | var q servingParkQuarantine |
| 123 | if boundary == "parent" && name == "namespace-parent.json" { |
| 124 | return unix.EIO |
| 125 | } |
| 126 | if name == "namespace-quarantine.json" && servingParkDecode(body, &q) == nil && (boundary == "fresh" && q.Fresh.Inode != 0 || boundary == "complete" && q.Complete) { |
| 127 | return unix.EIO |
| 128 | } |
| 129 | return nil |
| 130 | }} |
| 131 | if boundary == "quarantine-rename" { |
| 132 | fs.rename = func(a int, b string, c int, d string, flags uint) error { |
| 133 | if err := unix.Renameat2(a, b, c, d, flags); err != nil { |
| 134 | return err |
| 135 | } |
| 136 | return unix.EIO |
| 137 | } |
| 138 | } |
| 139 | if _, err := f.park(); !errors.Is(err, unix.EIO) { |
| 140 | t.Fatal("missing interrupted boundary", err) |
| 141 | } |
| 142 | before := f.f.read() |
| 143 | restarted := *f.p |
| 144 | ledger := *f.p.ledger |
| 145 | restarted.ledger = &ledger |
| 146 | freshFS := newServingParkFS(fs.data, f.disk) |
| 147 | freshFS.owner, freshFS.mount = fs.owner, fs.mount |
| 148 | ledger.fs = freshFS |
| 149 | restarted.root.Stop.Authenticate = func(context.Context, net.Conn) (StopPeer, error) { panic("startup used a peer") } |
| 150 | // Namespace recovery needs only the independently recorded parent, |
| 151 | // not a surviving source or a freshly selected host installation. |
| 152 | if err := os.Rename(filepath.Dir(f.db), filepath.Dir(f.db)+"-saved"); err != nil { |
| 153 | t.Fatal(err) |
| 154 | } |
| 155 | page, err := restarted.recoverServingParks(t.Context(), "", servingParkRecoveryLimit) |
| 156 | if err != nil || len(page.Results) != 0 || page.More { |
| 157 | t.Fatalf("startup recovery: %+v %v", page, err) |
| 158 | } |
| 159 | var q servingParkQuarantine |
| 160 | body, err := f.disk.Read("namespace-quarantine.json") |
| 161 | if err != nil || servingParkDecode(body, &q) != nil || !q.Complete { |
| 162 | t.Fatal("startup did not finish quarantine", err) |
| 163 | } |
| 164 | body, err = os.ReadFile(filepath.Join(filepath.Dir(path), q.name(), "unknown")) |
| 165 | if err != nil || string(body) != "retained" || !reflect.DeepEqual(before, f.f.read()) { |
| 166 | t.Fatal("startup changed retained bytes or claim", err) |
| 167 | } |
| 168 | beforeRecord, err := f.disk.Read("namespace-quarantine.json") |
| 169 | if err != nil { |
| 170 | t.Fatal(err) |
| 171 | } |
| 172 | if page, err = restarted.recoverServingParks(t.Context(), "", servingParkRecoveryLimit); err != nil || len(page.Results) != 0 { |
| 173 | t.Fatal("startup replay", page, err) |
| 174 | } |
| 175 | afterRecord, err := f.disk.Read("namespace-quarantine.json") |
| 176 | if err != nil || string(beforeRecord) != string(afterRecord) { |
| 177 | t.Fatal("completed startup rewrote quarantine", err) |
| 178 | } |
| 179 | }) |
| 180 | } |
| 181 | } |
| 182 | |
| 183 | func TestServingNamespaceStartupCannotCaptureParent(t *testing.T) { |
| 184 | f, _, path := quarantineFixture(t) |
| 185 | if err := f.ledger.withNamespace(t.Context(), func() error { return f.fs.recoverRecordedNamespace(t.Context()) }); err != nil { |
| 186 | t.Fatal(err) |
| 187 | } |
| 188 | if _, err := os.Stat(filepath.Join(path, "unknown")); err != nil { |
| 189 | t.Fatal("startup moved unrecorded obstruction", err) |
| 190 | } |
| 191 | for _, name := range []string{"namespace-parent.json", "namespace.json", "namespace-quarantine.json"} { |
| 192 | if _, err := f.disk.Read(name); !errors.Is(err, os.ErrNotExist) { |
| 193 | t.Fatal("startup created authority", name, err) |
| 194 | } |
| 195 | } |
| 196 | } |
| 197 | |
| 198 | func TestServingNamespaceParentCannotBeReplaced(t *testing.T) { |
| 199 | for _, fault := range []string{"parent", "corrupt-anchor", "corrupt-namespace", "mount", "mode"} { |
| 200 | t.Run(fault, func(t *testing.T) { |
| 201 | f := newServingRootFixture(t, true) |
| 202 | a, err := f.park() |
| 203 | if err != nil { |
| 204 | t.Fatal(err) |
| 205 | } |
| 206 | fs := f.p.ledger.fs.(*servingParkFS) |
| 207 | before, err := f.disk.Read("namespace-parent.json") |
| 208 | if err != nil { |
| 209 | t.Fatal(err) |
| 210 | } |
| 211 | switch fault { |
| 212 | case "parent": |
| 213 | path := filepath.Join(fs.data, "instances") |
| 214 | if err := os.Rename(path, path+"-saved"); err != nil { |
| 215 | t.Fatal(err) |
| 216 | } |
| 217 | if err := os.Mkdir(path, 0700); err != nil { |
| 218 | t.Fatal(err) |
| 219 | } |
| 220 | // Even moving the genuine inventoried child into the replacement |
| 221 | // parent does not authorize refreshing the recorded parent. |
| 222 | if err := os.Rename(filepath.Join(path+"-saved", provision.InstanceID(f.current.Host)), filepath.Dir(f.db)); err != nil { |
| 223 | t.Fatal(err) |
| 224 | } |
| 225 | case "corrupt-anchor": |
| 226 | if err := f.disk.Write("namespace-parent.json", []byte("{}")); err != nil { |
| 227 | t.Fatal(err) |
| 228 | } |
| 229 | before = []byte("{}") |
| 230 | case "corrupt-namespace": |
| 231 | if err := f.disk.Write("namespace.json", []byte("{}")); err != nil { |
| 232 | t.Fatal(err) |
| 233 | } |
| 234 | case "mount": |
| 235 | fs.mount = func(string) (bool, error) { return false, nil } |
| 236 | case "mode": |
| 237 | if err := os.Chmod(filepath.Join(fs.data, "instances"), 0777); err != nil { |
| 238 | t.Fatal(err) |
| 239 | } |
| 240 | } |
| 241 | page, err := f.p.recoverServingParks(t.Context(), "", servingParkRecoveryLimit) |
| 242 | if err != nil || len(page.Results) == 0 || page.Results[0].ID != "namespace" || page.Results[0].Err == nil { |
| 243 | t.Fatal("startup accepted unsafe parent", page, err) |
| 244 | } |
| 245 | if fault == "parent" { |
| 246 | if err := f.p.ledger.withNamespace(t.Context(), func() error { |
| 247 | return fs.recordNamespaceParent(t.Context(), f.current, a.Directory) |
| 248 | }); err == nil { |
| 249 | t.Fatal("parent recorder accepted a moved genuine source") |
| 250 | } |
| 251 | } |
| 252 | if _, err := f.park(); err == nil { |
| 253 | t.Fatal("authenticated request refreshed unsafe parent") |
| 254 | } |
| 255 | after, err := f.disk.Read("namespace-parent.json") |
| 256 | if err != nil || string(before) != string(after) { |
| 257 | t.Fatal("parent authority changed", err) |
| 258 | } |
| 259 | if _, err := f.disk.Read("namespace-quarantine.json"); !errors.Is(err, os.ErrNotExist) { |
| 260 | t.Fatal("unsafe parent authorized quarantine", err) |
| 261 | } |
| 262 | }) |
| 263 | } |
| 264 | } |
| 265 | |
| 266 | func TestServingNamespaceParentUpgradePreservesEarlierEvidence(t *testing.T) { |
| 267 | for _, fault := range []string{"", "parent", "corrupt-namespace", "corrupt-quarantine"} { |
| 268 | t.Run(fault, func(t *testing.T) { |
| 269 | f, parent, _ := quarantineFixture(t) |
| 270 | f.put("", "current database") |
| 271 | a, _ := f.evidence(10, true, "missing") |
| 272 | if _, err := quarantineRecover(t, f, parent); err != nil { |
| 273 | t.Fatal(err) |
| 274 | } |
| 275 | // These are checkpoint-14 namespace records: no parent anchor yet. |
| 276 | switch fault { |
| 277 | case "parent": |
| 278 | body, err := f.disk.Read("namespace.json") |
| 279 | if err != nil { |
| 280 | t.Fatal(err) |
| 281 | } |
| 282 | var n servingParkNamespace |
| 283 | if err := servingParkDecode(body, &n); err != nil { |
| 284 | t.Fatal(err) |
| 285 | } |
| 286 | n.Parent.Inode++ |
| 287 | if err := f.ledger.write("namespace.json", n); err != nil { |
| 288 | t.Fatal(err) |
| 289 | } |
| 290 | case "corrupt-namespace": |
| 291 | if err := f.disk.Write("namespace.json", []byte("{}")); err != nil { |
| 292 | t.Fatal(err) |
| 293 | } |
| 294 | case "corrupt-quarantine": |
| 295 | if err := f.disk.Write("namespace-quarantine.json", []byte("{}")); err != nil { |
| 296 | t.Fatal(err) |
| 297 | } |
| 298 | } |
| 299 | err := f.ledger.withNamespace(t.Context(), func() error { return f.fs.recordNamespaceParent(t.Context(), a.Successor, a.Directory) }) |
| 300 | if (err != nil) != (fault != "") { |
| 301 | t.Fatal("parent upgrade", err) |
| 302 | } |
| 303 | if fault != "" { |
| 304 | if _, err := f.disk.Read("namespace-parent.json"); !errors.Is(err, os.ErrNotExist) { |
| 305 | t.Fatal("upgrade blessed changed private evidence", err) |
| 306 | } |
| 307 | return |
| 308 | } |
| 309 | before, err := f.disk.Read("namespace-parent.json") |
| 310 | if err != nil { |
| 311 | t.Fatal(err) |
| 312 | } |
| 313 | // A failed write seam proves an established anchor is read-only. |
| 314 | f.fs.state = quarantineWriteFault{Store: f.disk, before: func(string, []byte) error { return unix.EIO }} |
| 315 | if err := f.ledger.withNamespace(t.Context(), func() error { |
| 316 | if err := f.fs.recordNamespaceParent(t.Context(), a.Successor, a.Directory); err != nil { |
| 317 | return err |
| 318 | } |
| 319 | return f.fs.recoverRecordedNamespace(t.Context()) |
| 320 | }); err != nil { |
| 321 | t.Fatal("completed recovery wrote state", err) |
| 322 | } |
| 323 | after, err := f.disk.Read("namespace-parent.json") |
| 324 | if err != nil || string(before) != string(after) { |
| 325 | t.Fatal("parent replay changed record", err) |
| 326 | } |
| 327 | }) |
| 328 | } |
| 329 | } |
| 330 | |