| 1 | package hostops |
| 2 | |
| 3 | import ( |
| 4 | "bytes" |
| 5 | "context" |
| 6 | "crypto/ed25519" |
| 7 | "crypto/sha256" |
| 8 | "encoding/hex" |
| 9 | "encoding/json" |
| 10 | "errors" |
| 11 | "fmt" |
| 12 | "os" |
| 13 | "sort" |
| 14 | |
| 15 | "github.com/carlosframework/platform/internal/hostactors" |
| 16 | "github.com/carlosframework/platform/internal/release" |
| 17 | ) |
| 18 | |
| 19 | // Publish freezes exact report bytes into #232's already-fenced backfill. |
| 20 | // The collecting manifest is the global create/record-writer freeze. Its ETag |
| 21 | // is supplied by the #232 coordinator, which retains that freeze through done. |
| 22 | func (b *Backfill) Publish(ctx context.Context, signer release.Signer, collectingETag string) error { |
| 23 | raw, tag, err := b.Store.Get(ctx, BackfillKey) |
| 24 | if err != nil { |
| 25 | return err |
| 26 | } |
| 27 | var p BackfillPlan |
| 28 | if tag != collectingETag || tag == "" || release.VerifyRootEnvelope(b.Directory.PublicKey, raw, &p) != nil || p.V != 1 || p.Phase != "collecting" || p.ID == "" || p.CreatedAt.IsZero() || p.Hosts == nil { |
| 29 | return ErrContract |
| 30 | } |
| 31 | a := ProductionAuthority{Store: b.Store, Directory: b.Directory} |
| 32 | census, err := a.census(ctx, "", "", p.ID, p.DirectoryRevision, nil) |
| 33 | if err != nil { |
| 34 | return err |
| 35 | } |
| 36 | p.Reports = map[string]string{} |
| 37 | for box, report := range census.Reports { |
| 38 | if report.CapturedAt.Before(p.CreatedAt) { |
| 39 | return ErrIdentity |
| 40 | } |
| 41 | key, err := reportKey(p.ID, box) |
| 42 | if err != nil { |
| 43 | return err |
| 44 | } |
| 45 | body, _, err := b.Store.Get(ctx, key) |
| 46 | if err != nil { |
| 47 | return err |
| 48 | } |
| 49 | var envelope CensusEnvelope |
| 50 | if json.Unmarshal(body, &envelope) != nil || envelope.Report.Reservation != report.Reservation || !envelope.Report.Complete { |
| 51 | return ErrIdentity |
| 52 | } |
| 53 | sum := sha256.Sum256(body) |
| 54 | p.Reports[box] = hex.EncodeToString(sum[:]) |
| 55 | } |
| 56 | // No activation over unfinished seed state (seeded-instances; merge |
| 57 | // review S2): with claims on nothing can commit, abort or finish it. |
| 58 | // Checked here, under the collecting freeze, on the census just read. |
| 59 | if err := activationSeedRefusal(census); err != nil { |
| 60 | return err |
| 61 | } |
| 62 | for _, r := range census.Records { |
| 63 | if err := CheckHostQuarantine(ctx, b.Store, r.Holder.Host); err != nil { |
| 64 | return err |
| 65 | } |
| 66 | selected, ok := p.Hosts[r.Holder.Host] |
| 67 | if !ok || selected.Key != r.Holder.Key || selected.ETag != r.ETag || selected.Acquisition == "" || r.Holder.Acquisition != selected.Acquisition { |
| 68 | return ErrIdentity |
| 69 | } |
| 70 | // Each host's box and region come from the signed census, never from |
| 71 | // a typed list: the one box whose report names the host, and that |
| 72 | // box's enrolled region. Adoption checks the same ownership again. |
| 73 | // A host no box reports keeps no box, so it is never adopted. A box |
| 74 | // or region the collecting plan already named must agree. |
| 75 | var owners []string |
| 76 | for ref, report := range census.Reports { |
| 77 | for _, i := range report.Installations { |
| 78 | if i.Host == r.Holder.Host { |
| 79 | owners = append(owners, ref) |
| 80 | } |
| 81 | } |
| 82 | } |
| 83 | if len(owners) > 1 { |
| 84 | return ErrIdentity |
| 85 | } |
| 86 | if len(owners) == 1 { |
| 87 | enrolled := census.Roster.Roster.Boxes[owners[0]] |
| 88 | if (selected.Box != "" && selected.Box != owners[0]) || (selected.Region != "" && selected.Region != enrolled.Region) || enrolled.Region == "" { |
| 89 | return ErrIdentity |
| 90 | } |
| 91 | selected.Box, selected.Region = owners[0], enrolled.Region |
| 92 | p.Hosts[r.Holder.Host] = selected |
| 93 | } |
| 94 | } |
| 95 | if len(p.Hosts) != len(census.Records) { |
| 96 | return ErrIdentity |
| 97 | } |
| 98 | p.Phase = "adopting" |
| 99 | body, err := release.SignRootEnvelope(signer, p) |
| 100 | if err != nil { |
| 101 | return err |
| 102 | } |
| 103 | _, err = b.Store.Put(ctx, BackfillKey, body, tag) |
| 104 | return err |
| 105 | } |
| 106 | |
| 107 | // Backfill.Done's refusals, one per condition. They are diagnostics: the |
| 108 | // claims run phase puts its own refusal in front of the ones an operator |
| 109 | // acts on (the plan, the records, the installations). |
| 110 | var ( |
| 111 | ErrDoneETag = fmt.Errorf("%w: backfill done: the plan's etag is not the collecting plan's", ErrContract) |
| 112 | ErrDoneEnvelope = fmt.Errorf("%w: backfill done: the plan does not verify", ErrContract) |
| 113 | ErrDonePhase = fmt.Errorf("%w: backfill done: the plan is not collecting", ErrContract) |
| 114 | ErrDoneRecords = fmt.Errorf("%w: backfill done: the record listing is not the plan's", ErrIdentity) |
| 115 | ErrDoneClaims = fmt.Errorf("%w: backfill done: control/hosts/ lists a claim", ErrIdentity) |
| 116 | ErrDoneAdmission = fmt.Errorf("%w: backfill done: an app admission exists", ErrIdentity) |
| 117 | ErrDoneReport = fmt.Errorf("%w: backfill done: a keyed box's census report does not verify", ErrIdentity) |
| 118 | ErrDoneStale = fmt.Errorf("%w: backfill done: a census report was captured before the plan", ErrIdentity) |
| 119 | ErrDoneInstallations = fmt.Errorf("%w: backfill done: a census report lists installations", ErrIdentity) |
| 120 | // ErrDoneRecordsChanged is Done's third listing failing after the write: |
| 121 | // done is written, over a record set that is no longer the plan's. It |
| 122 | // wraps ErrDoneRecords, so the claims run puts the records refusal in |
| 123 | // front of it. |
| 124 | ErrDoneRecordsChanged = fmt.Errorf("%w: done was written over a changed record set", ErrDoneRecords) |
| 125 | ) |
| 126 | |
| 127 | // Done closes an empty cohort's collecting plan straight to done (claim |
| 128 | // trial Option B, Paul's D1 of 2026-10-20). It adopts nothing, so it needs |
| 129 | // no report from a box that cannot report: what it proves is that every box |
| 130 | // that can report owns nothing, not that the others own everything else. |
| 131 | // |
| 132 | // Each of these is its own refusal, and a refusal writes nothing: |
| 133 | // - the plan's etag is collectingETag; |
| 134 | // - its root envelope verifies; |
| 135 | // - it is collecting; |
| 136 | // - the live record listing (Apps, Records) is the plan's hosts, by key, |
| 137 | // etag and acquisition, both before the reports are read and again |
| 138 | // immediately before the write; |
| 139 | // - control/hosts/ lists nothing; |
| 140 | // - no app has an admission.json; |
| 141 | // - every roster box with a census key has a report for the reservation |
| 142 | // that verifies exactly as the census verifies it, captured no earlier |
| 143 | // than the plan, listing no installation. |
| 144 | // |
| 145 | // Boxes with no census key are skipped and returned as unreported. The plan |
| 146 | // is then written as done under If-Match, with the reporting boxes' digests, |
| 147 | // read back, and CheckCreateFreeze must accept it. The records are listed a |
| 148 | // third time after that; a change there is ErrDoneRecordsChanged, with done |
| 149 | // already written. |
| 150 | func (b *Backfill) Done(ctx context.Context, signer release.Signer, collectingETag string) ([]string, error) { |
| 151 | raw, tag, err := b.Store.Get(ctx, BackfillKey) |
| 152 | if err != nil { |
| 153 | return nil, err |
| 154 | } |
| 155 | if tag == "" || tag != collectingETag { |
| 156 | return nil, ErrDoneETag |
| 157 | } |
| 158 | var p BackfillPlan |
| 159 | if release.VerifyRootEnvelope(b.Directory.PublicKey, raw, &p) != nil || p.V != 1 || !safeSegment(p.ID) || p.CreatedAt.IsZero() || p.Hosts == nil { |
| 160 | return nil, ErrDoneEnvelope |
| 161 | } |
| 162 | if p.Phase != "collecting" { |
| 163 | return nil, ErrDonePhase |
| 164 | } |
| 165 | roster, err := b.Directory.Read(ctx) |
| 166 | if err != nil { |
| 167 | return nil, err |
| 168 | } |
| 169 | if roster.Roster.Revision != p.DirectoryRevision { |
| 170 | return nil, ErrPending |
| 171 | } |
| 172 | |
| 173 | // The live record listing, as the census and collect list it. |
| 174 | apps, records, err := b.doneRecords(ctx, p.Hosts) |
| 175 | if err != nil { |
| 176 | return nil, err |
| 177 | } |
| 178 | |
| 179 | // No host claim, under the prefix the migration census lists. |
| 180 | claims, err := b.Store.List(ctx, "control/hosts/") |
| 181 | if err != nil { |
| 182 | return nil, err |
| 183 | } |
| 184 | if len(claims) != 0 { |
| 185 | return nil, ErrDoneClaims |
| 186 | } |
| 187 | |
| 188 | // No app admission (apps/<account>/<app>/admission.json). |
| 189 | for _, app := range apps { |
| 190 | key, err := AdmissionKey(app.Account, app.Name) |
| 191 | if err != nil { |
| 192 | return nil, err |
| 193 | } |
| 194 | if _, _, err := b.Store.Get(ctx, key); err == nil { |
| 195 | return nil, ErrDoneAdmission |
| 196 | } else if !errors.Is(err, os.ErrNotExist) { |
| 197 | return nil, err |
| 198 | } |
| 199 | } |
| 200 | |
| 201 | // Every keyed box reports, and reports nothing. |
| 202 | refs := make([]string, 0, len(roster.Roster.Boxes)) |
| 203 | for ref := range roster.Roster.Boxes { |
| 204 | refs = append(refs, ref) |
| 205 | } |
| 206 | sort.Strings(refs) |
| 207 | unreported := []string{} |
| 208 | p.Reports = map[string]string{} |
| 209 | for _, ref := range refs { |
| 210 | box := roster.Roster.Boxes[ref] |
| 211 | if len(box.CensusPublicKey) == 0 { |
| 212 | unreported = append(unreported, ref) |
| 213 | continue |
| 214 | } |
| 215 | key, err := reportKey(p.ID, ref) |
| 216 | if err != nil { |
| 217 | return nil, err |
| 218 | } |
| 219 | body, reportTag, err := b.Store.Get(ctx, key) |
| 220 | if err != nil { |
| 221 | return nil, errors.Join(ErrDoneReport, err) |
| 222 | } |
| 223 | report, err := verifyCensusReport(body, reportTag, ref, box, p.ID, roster.Roster.Revision, records) |
| 224 | if err != nil { |
| 225 | return nil, errors.Join(ErrDoneReport, err) |
| 226 | } |
| 227 | if report.CapturedAt.Before(p.CreatedAt) { |
| 228 | return nil, ErrDoneStale |
| 229 | } |
| 230 | if len(report.Installations) != 0 { |
| 231 | return nil, ErrDoneInstallations |
| 232 | } |
| 233 | sum := sha256.Sum256(body) |
| 234 | p.Reports[ref] = hex.EncodeToString(sum[:]) |
| 235 | } |
| 236 | // A done with no report at all would prove nothing. |
| 237 | if len(p.Reports) == 0 { |
| 238 | return nil, ErrDoneReport |
| 239 | } |
| 240 | fresh, err := b.Directory.Read(ctx) |
| 241 | if err != nil { |
| 242 | return nil, err |
| 243 | } |
| 244 | if fresh.ETag != roster.ETag { |
| 245 | return nil, ErrPending |
| 246 | } |
| 247 | |
| 248 | p.Phase = "done" |
| 249 | body, err := release.SignRootEnvelope(signer, p) |
| 250 | if err != nil { |
| 251 | return nil, err |
| 252 | } |
| 253 | // The second listing, after every report was fetched and verified and |
| 254 | // immediately before the write: a record created while the reports were |
| 255 | // read is refused here, and nothing is written. The plan's If-Match |
| 256 | // covers only the plan, so this narrows the window; the trial closes it |
| 257 | // by stopping the console's record writes for the window. |
| 258 | if _, _, err := b.doneRecords(ctx, p.Hosts); err != nil { |
| 259 | return nil, err |
| 260 | } |
| 261 | if _, err = b.Store.Put(ctx, BackfillKey, body, tag); err != nil { |
| 262 | return nil, err |
| 263 | } |
| 264 | got, _, err := b.Store.Get(ctx, BackfillKey) |
| 265 | if err != nil { |
| 266 | return nil, err |
| 267 | } |
| 268 | if !bytes.Equal(got, body) { |
| 269 | return nil, ErrContract |
| 270 | } |
| 271 | if err := CheckCreateFreeze(ctx, b.Store, b.Directory.PublicKey); err != nil { |
| 272 | return nil, err |
| 273 | } |
| 274 | // The third listing, after the write and its readback: a record that |
| 275 | // appeared between the second listing and now is not in the done plan, |
| 276 | // so done was written over a changed record set. Say so; the write |
| 277 | // cannot be taken back here. |
| 278 | if _, _, err := b.doneRecords(ctx, p.Hosts); err != nil { |
| 279 | if errors.Is(err, ErrDoneRecords) { |
| 280 | return nil, ErrDoneRecordsChanged |
| 281 | } |
| 282 | return nil, err |
| 283 | } |
| 284 | return unreported, nil |
| 285 | } |
| 286 | |
| 287 | // doneRecords lists the live records as collect and the census list them |
| 288 | // (Apps, then Records per app) and refuses, with ErrDoneRecords, unless |
| 289 | // they are hosts by key, etag and acquisition, no more and no fewer. Done |
| 290 | // runs it before the reports, again immediately before its write and once |
| 291 | // more after it. |
| 292 | func (b *Backfill) doneRecords(ctx context.Context, hosts map[string]BackfillHost) ([]App, map[string]Record, error) { |
| 293 | apps, err := Apps(ctx, b.Store) |
| 294 | if err != nil { |
| 295 | return nil, nil, err |
| 296 | } |
| 297 | records := map[string]Record{} |
| 298 | for _, app := range apps { |
| 299 | rs, err := Records(ctx, b.Store, app.Account, app.Name) |
| 300 | if err != nil { |
| 301 | return nil, nil, err |
| 302 | } |
| 303 | for _, r := range rs { |
| 304 | if _, dup := records[r.Holder.Host]; dup { |
| 305 | return nil, nil, ErrDoneRecords |
| 306 | } |
| 307 | records[r.Holder.Host] = r |
| 308 | } |
| 309 | } |
| 310 | if len(records) != len(hosts) { |
| 311 | return nil, nil, ErrDoneRecords |
| 312 | } |
| 313 | for host, r := range records { |
| 314 | frozen, ok := hosts[host] |
| 315 | if !ok || frozen.Key != r.Holder.Key || frozen.ETag != r.ETag || frozen.Acquisition == "" || frozen.Acquisition != r.Holder.Acquisition { |
| 316 | return nil, nil, ErrDoneRecords |
| 317 | } |
| 318 | } |
| 319 | return apps, records, nil |
| 320 | } |
| 321 | |
| 322 | // verifyCensusReport is the census's acceptance of one roster box's report |
| 323 | // for the reservation, for Backfill.Done: the envelope verifies under the |
| 324 | // box's enrolled census key, the report is complete and names this box, |
| 325 | // enrollment, reservation and roster revision, and every installation names |
| 326 | // a listed record, once. It is ProductionAuthority.census's per-box check |
| 327 | // line for line; census is an accepted-legacy unit and is left untouched, |
| 328 | // and TestDoneVerifiesReportsAsTheCensusDoes holds the two together. |
| 329 | func verifyCensusReport(raw []byte, tag, ref string, box hostactors.Box, reservation string, revision uint64, records map[string]Record) (CensusReport, error) { |
| 330 | var envelope CensusEnvelope |
| 331 | if decode(raw, &envelope) != nil { |
| 332 | return CensusReport{}, ErrIdentity |
| 333 | } |
| 334 | report := envelope.Report |
| 335 | canonical, e := json.Marshal(report) |
| 336 | if e != nil || len(box.CensusPublicKey) != ed25519.PublicKeySize || !ed25519.Verify(box.CensusPublicKey, canonical, envelope.Signature) { |
| 337 | return CensusReport{}, ErrIdentity |
| 338 | } |
| 339 | if tag == "" || !report.Complete || report.Box != ref || report.Enrollment != box.Enrollment || report.Reservation != reservation || report.DirectoryRevision != revision || report.BootID == "" || report.CapturedAt.IsZero() || report.Installations == nil { |
| 340 | return CensusReport{}, ErrIdentity |
| 341 | } |
| 342 | seen := map[string]bool{} |
| 343 | for _, i := range report.Installations { |
| 344 | if i.Host == "" || seen[i.Host] || (!i.Route && !i.Local) { |
| 345 | return CensusReport{}, ErrIdentity |
| 346 | } |
| 347 | seen[i.Host] = true |
| 348 | r, ok := records[i.Host] |
| 349 | if !ok || i.Account != r.Holder.Account || i.App != r.Holder.App || i.Key != r.Holder.Key || i.Family != r.Holder.Family { |
| 350 | return CensusReport{}, ErrIdentity |
| 351 | } |
| 352 | inc, err := r.Incarnation() |
| 353 | if err != nil || (i.Incarnation != "" && i.Incarnation != inc) { |
| 354 | return CensusReport{}, ErrIdentity |
| 355 | } |
| 356 | } |
| 357 | return report, nil |
| 358 | } |
| 359 | |
| 360 | // VerifyAdopting is the adopters' own acceptance of a published plan, run |
| 361 | // by the signer as readback after Publish: the signed plan reads adopting, |
| 362 | // and the complete census it froze verifies against its report digests, as |
| 363 | // Adoption and AuthorizeBackfillApp require. It writes nothing. |
| 364 | func (b *Backfill) VerifyAdopting(ctx context.Context) (BackfillPlan, error) { |
| 365 | p, tag, err := b.read(ctx) |
| 366 | if err != nil { |
| 367 | return p, err |
| 368 | } |
| 369 | a := ProductionAuthority{Store: b.Store, Directory: b.Directory} |
| 370 | if _, err = a.census(ctx, "", "", p.ID, p.DirectoryRevision, p.Reports); err != nil { |
| 371 | return p, err |
| 372 | } |
| 373 | _, fresh, err := b.read(ctx) |
| 374 | if err != nil { |
| 375 | return p, err |
| 376 | } |
| 377 | if fresh != tag { |
| 378 | return p, ErrPending |
| 379 | } |
| 380 | return p, nil |
| 381 | } |
| 382 | |
| 383 | // PublishCensusReport is called by the guarded box census collector. Each box |
| 384 | // owns its private signing key; the root-signed roster binds its public key to |
| 385 | // verified enrollment. The console has only that public key. |
| 386 | func PublishCensusReport(ctx context.Context, store release.Store, d hostactors.Directory, report CensusReport, private ed25519.PrivateKey) error { |
| 387 | if len(private) != ed25519.PrivateKeySize { |
| 388 | return ErrContract |
| 389 | } |
| 390 | roster, err := d.Read(ctx) |
| 391 | if err != nil { |
| 392 | return err |
| 393 | } |
| 394 | box, ok := roster.Roster.Boxes[report.Box] |
| 395 | if !ok || report.Enrollment != box.Enrollment || report.DirectoryRevision != roster.Roster.Revision || !bytes.Equal(private.Public().(ed25519.PublicKey), box.CensusPublicKey) || report.BootID == "" || !report.Complete || report.Installations == nil || report.CapturedAt.IsZero() { |
| 396 | return ErrIdentity |
| 397 | } |
| 398 | canonical, err := json.Marshal(report) |
| 399 | if err != nil { |
| 400 | return err |
| 401 | } |
| 402 | raw, err := json.Marshal(CensusEnvelope{Report: report, Signature: ed25519.Sign(private, canonical)}) |
| 403 | if err != nil { |
| 404 | return err |
| 405 | } |
| 406 | key, err := reportKey(report.Reservation, report.Box) |
| 407 | if err != nil { |
| 408 | return err |
| 409 | } |
| 410 | if _, err = store.Put(ctx, key, raw, ""); err == nil { |
| 411 | return nil |
| 412 | } |
| 413 | if !errors.Is(err, release.ErrCASConflict) { |
| 414 | return err |
| 415 | } |
| 416 | previous, _, err := store.Get(ctx, key) |
| 417 | if err != nil { |
| 418 | return err |
| 419 | } |
| 420 | if !bytes.Equal(previous, raw) { |
| 421 | return ErrIdentity |
| 422 | } |
| 423 | return nil |
| 424 | } |
| 425 | |
| 426 | // AuthorizeBackfillApp is the only legacy app-admission initialization proof. |
| 427 | // It accepts the same complete frozen census as host adoption, including apps |
| 428 | // with no host records. Empty results from a failed listing never qualify. |
| 429 | func (a ProductionAuthority) AuthorizeBackfillApp(ctx context.Context, account, app string) (string, error) { |
| 430 | if a.Backfill == nil || !safeSegment(account) || !safeSegment(app) { |
| 431 | return "", ErrContract |
| 432 | } |
| 433 | p, tag, err := a.Backfill.read(ctx) |
| 434 | if err != nil { |
| 435 | return "", err |
| 436 | } |
| 437 | census, err := a.census(ctx, account, app, p.ID, p.DirectoryRevision, p.Reports) |
| 438 | if err != nil { |
| 439 | return "", err |
| 440 | } |
| 441 | apps, err := Apps(ctx, a.Store) |
| 442 | if err != nil { |
| 443 | return "", err |
| 444 | } |
| 445 | found := false |
| 446 | for _, candidate := range apps { |
| 447 | if candidate.Account == account && candidate.Name == app { |
| 448 | found = true |
| 449 | } |
| 450 | } |
| 451 | if !found { |
| 452 | return "", ErrIdentity |
| 453 | } |
| 454 | for _, r := range census.Records { |
| 455 | planned, ok := p.Hosts[r.Holder.Host] |
| 456 | if !ok || planned.Key != r.Holder.Key || planned.ETag != r.ETag || planned.Acquisition == "" || planned.Acquisition != r.Holder.Acquisition { |
| 457 | return "", ErrIdentity |
| 458 | } |
| 459 | if err = CheckHostQuarantine(ctx, a.Store, r.Holder.Host); err != nil { |
| 460 | return "", err |
| 461 | } |
| 462 | } |
| 463 | _, fresh, err := a.Backfill.read(ctx) |
| 464 | if err != nil { |
| 465 | return "", err |
| 466 | } |
| 467 | if fresh != tag { |
| 468 | return "", ErrPending |
| 469 | } |
| 470 | return p.ID, nil |
| 471 | } |
| 472 | |
amadan