| 1 | //go:build linux |
| 2 | |
| 3 | package host |
| 4 | |
| 5 | import ( |
| 6 | "errors" |
| 7 | "fmt" |
| 8 | "os" |
| 9 | "reflect" |
| 10 | "testing" |
| 11 | |
| 12 | "github.com/carlosframework/platform/internal/hostbox" |
| 13 | ) |
| 14 | |
| 15 | // Put the historical records on a different boot-local device number without |
| 16 | // changing any persistent proof or anything in the actual filesystem. |
| 17 | func servingRenumberHistory(t *testing.T, f *servingParkDiskFixture, r *servingParkRecord) { |
| 18 | t.Helper() |
| 19 | old := r.Source.Parent.Device + 123 |
| 20 | for _, d := range []*servingArtifactDirectory{&r.Source.Parent, &r.Source.Directory, &r.Authorization.Directory, &r.Namespace, &r.Event} { |
| 21 | if d.Inode != 0 { |
| 22 | d.Device = old |
| 23 | } |
| 24 | } |
| 25 | for i := range r.Source.Files { |
| 26 | if r.Source.Files[i].Present { |
| 27 | r.Source.Files[i].Device = old |
| 28 | } |
| 29 | } |
| 30 | if err := f.ledger.save(*r); err != nil { |
| 31 | t.Fatal(err) |
| 32 | } |
| 33 | p, err := f.fs.historicalNamespaceParent() |
| 34 | if err != nil { |
| 35 | t.Fatal(err) |
| 36 | } |
| 37 | p.Parent.Device = old |
| 38 | if err := f.ledger.write("namespace-parent.json", p); err != nil { |
| 39 | t.Fatal(err) |
| 40 | } |
| 41 | var n servingParkNamespace |
| 42 | body, err := f.disk.Read("namespace.json") |
| 43 | if err != nil || servingParkDecode(body, &n) != nil { |
| 44 | t.Fatal(err) |
| 45 | } |
| 46 | n.Parent.Device, n.Root.Device = old, old |
| 47 | if err := f.ledger.write("namespace.json", n); err != nil { |
| 48 | t.Fatal(err) |
| 49 | } |
| 50 | if r.Authorization.Legacy { |
| 51 | c, err := f.ledger.legacyConsumption(r.Authorization.Successor) |
| 52 | if err != nil { |
| 53 | t.Fatal(err) |
| 54 | } |
| 55 | if c == nil { |
| 56 | return |
| 57 | } |
| 58 | c.Source, c.Authorization = r.Source, r.Authorization |
| 59 | if err := f.ledger.write(servingLegacyKey("legacy", r.Authorization.Successor), c); err != nil { |
| 60 | t.Fatal(err) |
| 61 | } |
| 62 | } |
| 63 | } |
| 64 | |
| 65 | func TestServingParkRebootDeviceRenumber(t *testing.T) { |
| 66 | for _, phase := range []string{"captured", "prepared", "rename-reply", "parked", "complete", "unlink-reply", "partial-unlink", "rmdir-reply", "closed"} { |
| 67 | t.Run(phase, func(t *testing.T) { |
| 68 | f, r := servingRebootRecordFixture(t, phase) |
| 69 | servingRenumberHistory(t, f, &r) |
| 70 | before := servingBootSnapshot(t, f.disk.dir) |
| 71 | got, err := f.fs.inspectRebootRecord(t.Context(), r.Authorization.ID) |
| 72 | if err != nil || got.Topology.CurrentParent.Device == r.Source.Parent.Device || !reflect.DeepEqual(got.Record, r) { |
| 73 | t.Fatal("renumbered-device inspection", got, err) |
| 74 | } |
| 75 | if !reflect.DeepEqual(before, servingBootSnapshot(t, f.disk.dir)) { |
| 76 | t.Fatal("inspection rewrote history") |
| 77 | } |
| 78 | if err := f.fs.CheckParked(&r); err == nil { |
| 79 | t.Fatal("inspection granted old-boot authority") |
| 80 | } |
| 81 | if phase == "captured" || phase == "prepared" || phase == "rename-reply" { |
| 82 | return |
| 83 | } |
| 84 | var active *servingParkRebootAuthority |
| 85 | if !r.Complete { |
| 86 | a := r.activeAuthorization() |
| 87 | active = &servingParkRebootAuthority{ClaimETag: "fixture", Actor: hostbox.Actor{Box: a.Successor.Box, Region: a.Successor.Region, Enrollment: "fixture", Revision: 1, Binding: "1", Eligible: true}, Directory: got.Source, Tags: a.TagsFrom} |
| 88 | } |
| 89 | if err := f.ledger.prepareRebootLocked(t.Context(), r, active); err != nil { |
| 90 | t.Fatal("prepare", err) |
| 91 | } |
| 92 | if err := f.ledger.commitRebootLocked(t.Context(), r, active); err != nil { |
| 93 | t.Fatal("commit", err) |
| 94 | } |
| 95 | current, err := f.ledger.read(r.Authorization.ID) |
| 96 | if err != nil || current.Source.Parent.Device != got.Topology.CurrentParent.Device { |
| 97 | t.Fatal("current device", current, err) |
| 98 | } |
| 99 | if err := f.ledger.save(current); err != nil { |
| 100 | t.Fatal("reverse binding", err) |
| 101 | } |
| 102 | raw, err := f.ledger.readHistorical(r.Authorization.ID) |
| 103 | if err != nil || !reflect.DeepEqual(raw, r) { |
| 104 | t.Fatal("round trip changed history", raw, err) |
| 105 | } |
| 106 | if r.Parked && !r.Deleting { |
| 107 | if err := f.fs.CheckParked(¤t); !r.Complete && err != nil { |
| 108 | t.Fatal("bound park", err) |
| 109 | } |
| 110 | if r.Complete { |
| 111 | f.now = current.ReapAfter |
| 112 | if err := f.ledger.reapLocked(t.Context(), r.Authorization.ID); err != nil { |
| 113 | t.Fatal("reap", err) |
| 114 | } |
| 115 | if _, err := os.Stat(f.fs.data + "/instances/.serving-parked/" + r.Authorization.ID); !os.IsNotExist(err) { |
| 116 | t.Fatal("event survived reap", err) |
| 117 | } |
| 118 | } |
| 119 | } |
| 120 | }) |
| 121 | } |
| 122 | } |
| 123 | |
| 124 | func TestServingParkRebootDeviceRenumberRefusesReplacement(t *testing.T) { |
| 125 | for _, fault := range []string{"filesystem", "root-handle", "source-handle", "event-handle", "file-handle", "inode", "same-boot"} { |
| 126 | t.Run(fault, func(t *testing.T) { |
| 127 | f, r := servingRebootRecordFixture(t, "parked") |
| 128 | servingRenumberHistory(t, f, &r) |
| 129 | var n servingParkNamespace |
| 130 | body, err := f.disk.Read("namespace.json") |
| 131 | if err != nil || servingParkDecode(body, &n) != nil { |
| 132 | t.Fatal(err) |
| 133 | } |
| 134 | switch fault { |
| 135 | case "filesystem": |
| 136 | n.Topology.Parent.Filesystem[0]++ |
| 137 | n.Topology.Root.Filesystem = n.Topology.Parent.Filesystem |
| 138 | case "root-handle": |
| 139 | n.Topology.Root.Handle = "aa" |
| 140 | case "source-handle": |
| 141 | r.Source.Persistent.Handle = "aa" |
| 142 | case "event-handle": |
| 143 | r.EventPersistent.Handle = "aa" |
| 144 | case "file-handle": |
| 145 | r.Source.Files[0].Persistent.Handle = "aa" |
| 146 | case "inode": |
| 147 | r.Source.Directory.Inode++ |
| 148 | r.Authorization.Directory = r.Source.Directory |
| 149 | case "same-boot": |
| 150 | f.fs.boot = func() (string, error) { return r.Source.BootID, nil } |
| 151 | } |
| 152 | if err := f.ledger.save(r); err != nil { |
| 153 | t.Fatal(err) |
| 154 | } |
| 155 | if err := f.ledger.write("namespace.json", n); err != nil { |
| 156 | t.Fatal(err) |
| 157 | } |
| 158 | before := servingBootSnapshot(t, f.disk.dir) |
| 159 | if _, err := f.fs.inspectRebootRecord(t.Context(), r.Authorization.ID); err == nil { |
| 160 | t.Fatal("renumbering accepted replaced identity") |
| 161 | } |
| 162 | if !reflect.DeepEqual(before, servingBootSnapshot(t, f.disk.dir)) { |
| 163 | t.Fatal("refusal changed authority") |
| 164 | } |
| 165 | }) |
| 166 | } |
| 167 | } |
| 168 | |
| 169 | func TestServingParkRebootDeviceRenumberLostWrite(t *testing.T) { |
| 170 | for _, before := range []bool{true, false} { |
| 171 | for at := 1; at <= 3; at++ { |
| 172 | f, r := servingRebootRecordFixture(t, "complete") |
| 173 | servingRenumberHistory(t, f, &r) |
| 174 | fault := &servingParkFaultStore{Store: f.disk, failAt: at, before: before} |
| 175 | f.ledger.state, f.fs.state = fault, fault |
| 176 | root := &servingParkRoot{ledger: f.ledger} |
| 177 | err := root.commitServingReboot(t.Context(), r.Authorization.ID) |
| 178 | if !errors.Is(err, os.ErrPermission) { |
| 179 | t.Fatalf("%v/%d: missing write fault: %v", before, at, err) |
| 180 | } |
| 181 | f.ledger.state, f.fs.state = f.disk, f.disk |
| 182 | if err := root.commitServingReboot(t.Context(), r.Authorization.ID); err != nil { |
| 183 | t.Fatal("commit replay", err) |
| 184 | } |
| 185 | if err := f.fs.recoverRecordedNamespace(t.Context()); err != nil { |
| 186 | t.Fatal("namespace recovery", err) |
| 187 | } |
| 188 | current, err := f.ledger.read(r.Authorization.ID) |
| 189 | if err != nil { |
| 190 | t.Fatal(err) |
| 191 | } |
| 192 | f.now = current.ReapAfter |
| 193 | if err := f.ledger.reapLocked(t.Context(), r.Authorization.ID); err != nil { |
| 194 | t.Fatal("reap after lost reply", err) |
| 195 | } |
| 196 | raw, err := f.ledger.readHistorical(r.Authorization.ID) |
| 197 | if err != nil || raw.Source.BootID != r.Source.BootID || raw.Source.Parent != r.Source.Parent || !raw.Closed { |
| 198 | t.Fatal("history after recovery", raw, err) |
| 199 | } |
| 200 | } |
| 201 | } |
| 202 | } |
| 203 | |
| 204 | func TestServingParkRetirementAfterDeviceRenumber(t *testing.T) { |
| 205 | for _, renumberAgain := range []bool{false, true} { |
| 206 | for _, replacement := range []bool{false, true} { |
| 207 | t.Run(fmt.Sprint(renumberAgain, "/replacement-", replacement), func(t *testing.T) { |
| 208 | f, old := servingRebootRecordFixture(t, "complete") |
| 209 | servingRenumberHistory(t, f, &old) // A/D1 -> B/D2. |
| 210 | root := &servingParkRoot{ledger: f.ledger} |
| 211 | if err := root.commitServingReboot(t.Context(), old.Authorization.ID); err != nil { |
| 212 | t.Fatal(err) |
| 213 | } |
| 214 | f.put("", "boot B served database") |
| 215 | f.put(".gen", "G2") |
| 216 | a, stops := f.evidence(2, false, "started") |
| 217 | park, err := f.park(a, stops) |
| 218 | if err != nil { |
| 219 | t.Fatal("B retirement and replacement", err) |
| 220 | } |
| 221 | f.complete(a) |
| 222 | park, err = f.ledger.readHistorical(a.ID) |
| 223 | if err != nil { |
| 224 | t.Fatal(err) |
| 225 | } |
| 226 | if renumberAgain { |
| 227 | // C's real device is D3; B's immutable record used D2. |
| 228 | b := servingParkBootBinding{Version: 1, ID: "retirement-" + a.ID, BeforeBoot: park.Source.BootID, BootID: park.Source.BootID, BeforeParent: park.Source.Parent, Parent: park.Source.Parent} |
| 229 | b.Parent.Device += 456 |
| 230 | if err := b.record(&park, false); err != nil { |
| 231 | t.Fatal(err) |
| 232 | } |
| 233 | if err := f.ledger.save(park); err != nil { |
| 234 | t.Fatal(err) |
| 235 | } |
| 236 | } |
| 237 | f.fs.boot = func() (string, error) { return "33333333-3333-4333-8333-333333333333", nil } |
| 238 | if err := root.commitServingReboot(t.Context(), a.ID); err != nil { |
| 239 | t.Fatal("C recovery", err) |
| 240 | } |
| 241 | if replacement { |
| 242 | f.put("", "boot C served database") |
| 243 | f.put(".gen", "G3") |
| 244 | next, stops := f.evidence(3, false, "started") |
| 245 | if _, err := f.park(next, stops); err != nil { |
| 246 | t.Fatal("C replacement", err) |
| 247 | } |
| 248 | f.complete(next) |
| 249 | nextPark, err := f.ledger.read(next.ID) |
| 250 | if err != nil { |
| 251 | t.Fatal(err) |
| 252 | } |
| 253 | f.now = nextPark.ReapAfter |
| 254 | if err := f.ledger.reapLocked(t.Context(), next.ID); err != nil { |
| 255 | t.Fatal("C replacement reaping", err) |
| 256 | } |
| 257 | } |
| 258 | f.now = park.ReapAfter |
| 259 | if err := f.ledger.reapLocked(t.Context(), a.ID); err != nil { |
| 260 | t.Fatal("C reap", err) |
| 261 | } |
| 262 | if _, err := os.Stat(f.path(a, "")); !os.IsNotExist(err) { |
| 263 | t.Fatal("park survived", err) |
| 264 | } |
| 265 | }) |
| 266 | } |
| 267 | } |
| 268 | |
| 269 | } |
| 270 | |
| 271 | func TestServingParkLaterRetirementRequiresParentProof(t *testing.T) { |
| 272 | for _, fault := range []string{"missing", "handle", "filesystem"} { |
| 273 | t.Run(fault, func(t *testing.T) { |
| 274 | f, r := servingRebootRecordFixture(t, "complete") |
| 275 | servingRenumberHistory(t, f, &r) |
| 276 | root := &servingParkRoot{ledger: f.ledger} |
| 277 | if err := root.commitServingReboot(t.Context(), r.Authorization.ID); err != nil { |
| 278 | t.Fatal(err) |
| 279 | } |
| 280 | f.put("", "later served database") |
| 281 | f.put(".gen", "G2") |
| 282 | a, stops := f.evidence(2, false, "started") |
| 283 | if _, err := f.park(a, stops); err != nil { |
| 284 | t.Fatal(err) |
| 285 | } |
| 286 | f.complete(a) |
| 287 | later, err := f.ledger.readHistorical(a.ID) |
| 288 | if err != nil { |
| 289 | t.Fatal(err) |
| 290 | } |
| 291 | switch fault { |
| 292 | case "missing": |
| 293 | later.Source.ParentPersistent = servingParkPersistentDirectory{} |
| 294 | case "handle": |
| 295 | later.Source.ParentPersistent.Handle = "aa" |
| 296 | case "filesystem": |
| 297 | later.Source.ParentPersistent.Filesystem[0]++ |
| 298 | } |
| 299 | if err := f.ledger.write("retirement-"+a.ID+".json", later); err != nil { |
| 300 | t.Fatal(err) |
| 301 | } |
| 302 | f.fs.boot = func() (string, error) { return "33333333-3333-4333-8333-333333333333", nil } |
| 303 | before := servingBootSnapshot(t, f.disk.dir) |
| 304 | if err := root.commitServingReboot(t.Context(), a.ID); err == nil { |
| 305 | t.Fatal("unbound historical parent accepted") |
| 306 | } |
| 307 | if !reflect.DeepEqual(before, servingBootSnapshot(t, f.disk.dir)) { |
| 308 | t.Fatal("refusal rewrote authority") |
| 309 | } |
| 310 | f.retained(a, "later served database", later.ReapAfter) |
| 311 | }) |
| 312 | } |
| 313 | } |
| 314 | |