| 1 | package hostops |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "crypto/ed25519" |
| 6 | "crypto/sha256" |
| 7 | "encoding/hex" |
| 8 | "encoding/json" |
| 9 | "errors" |
| 10 | "os" |
| 11 | "sort" |
| 12 | "sync" |
| 13 | "time" |
| 14 | |
| 15 | "github.com/carlosframework/platform/internal/domains" |
| 16 | "github.com/carlosframework/platform/internal/hostactors" |
| 17 | "github.com/carlosframework/platform/internal/hostclaim" |
| 18 | "github.com/carlosframework/platform/internal/provision" |
| 19 | "github.com/carlosframework/platform/internal/release" |
| 20 | ) |
| 21 | |
| 22 | const BackfillKey = "control/hosts-backfill.json" |
| 23 | const HostsQuarantineKey = "control/hosts-quarantine.json" |
| 24 | |
| 25 | type CensusInstallation struct { |
| 26 | Placement string `json:"placement"` |
| 27 | Host string `json:"host"` |
| 28 | Account string `json:"account"` |
| 29 | App string `json:"app"` |
| 30 | Family string `json:"family"` |
| 31 | Key string `json:"key"` |
| 32 | Incarnation string `json:"incarnation"` |
| 33 | Stamp uint64 `json:"stamp"` |
| 34 | Local bool `json:"local"` |
| 35 | Route bool `json:"route"` |
| 36 | } |
| 37 | type CensusReport struct { |
| 38 | Box string `json:"box"` |
| 39 | Enrollment string `json:"enrollment"` |
| 40 | BootID string `json:"boot_id"` |
| 41 | Reservation string `json:"reservation"` |
| 42 | DirectoryRevision uint64 `json:"directory_revision"` |
| 43 | CapturedAt time.Time `json:"captured_at"` |
| 44 | Complete bool `json:"complete"` |
| 45 | Installations []CensusInstallation `json:"installations"` |
| 46 | // Seeds is the box's own seed state (backfill_seeds.go). Signed with |
| 47 | // the rest of the report. Nil is a box that did not say, which |
| 48 | // activation reads as unknown and refuses. |
| 49 | Seeds *CensusSeeds `json:"seeds,omitempty"` |
| 50 | } |
| 51 | type Census struct { |
| 52 | Records []Record |
| 53 | Reports map[string]CensusReport |
| 54 | Roster hostactors.Observed |
| 55 | } |
| 56 | type BackfillHost struct { |
| 57 | Key string `json:"key"` |
| 58 | ETag string `json:"etag"` |
| 59 | Acquisition string `json:"acq"` |
| 60 | Region string `json:"region"` |
| 61 | Box string `json:"box"` |
| 62 | } |
| 63 | type BackfillPlan struct { |
| 64 | V int `json:"v"` |
| 65 | ID string `json:"id"` |
| 66 | Phase string `json:"phase"` |
| 67 | DirectoryRevision uint64 `json:"directory_revision"` |
| 68 | CreatedAt time.Time `json:"created_at"` |
| 69 | Reports map[string]string `json:"reports"` |
| 70 | Hosts map[string]BackfillHost `json:"hosts"` |
| 71 | } |
| 72 | type HostsQuarantine struct { |
| 73 | V int `json:"v"` |
| 74 | Hosts map[string]string `json:"hosts"` |
| 75 | } |
| 76 | type CensusEnvelope struct { |
| 77 | Report CensusReport `json:"report"` |
| 78 | Signature []byte `json:"signature"` |
| 79 | } |
| 80 | |
| 81 | type Backfill struct { |
| 82 | mu sync.Mutex |
| 83 | cached *Census |
| 84 | cachedTag string |
| 85 | Store release.Store |
| 86 | Directory hostactors.Directory |
| 87 | } |
| 88 | type ProductionAuthority struct { |
| 89 | HomeRegion string |
| 90 | Store release.Store |
| 91 | Control *ControlWriter |
| 92 | Directory hostactors.Directory |
| 93 | Now func() time.Time |
| 94 | ConsoleID, Enrollment, Namespace string |
| 95 | Signer release.Signer |
| 96 | Evidence hostactors.Evidence |
| 97 | Backfill *Backfill |
| 98 | // Executors is create's executor selection (region HA §5.2, L8). The |
| 99 | // create door sets it only with CARLOS_REGION_HA on; nil keeps today's |
| 100 | // create path (claim Plan 5's selectExecutor), so a claim-enabled |
| 101 | // deployment with HA off (every one-box region included) creates |
| 102 | // exactly as before. |
| 103 | Executors *ExecutorChooser |
| 104 | } |
| 105 | |
| 106 | // SelectExecutor is Authority's create-time selection: choose() once, |
| 107 | // before the create CAS, when a chooser is configured; otherwise no |
| 108 | // selection, and the create is today's. The chooser chooses within claim |
| 109 | // Plan 5's scope: this authority's namespace and the record's pool. |
| 110 | // |
| 111 | // A door that already chose for this host (WithChosenExecutor) is held to |
| 112 | // that choice, so the create names the box the door checked (H27). |
| 113 | func (a ProductionAuthority) SelectExecutor(ctx context.Context, obs hostclaim.Observed, r Record) (hostclaim.ExecutorSelection, bool, error) { |
| 114 | if a.Executors == nil { |
| 115 | return hostclaim.ExecutorSelection{}, false, nil |
| 116 | } |
| 117 | if sel, ok := ChosenExecutor(ctx, r.Holder.Host); ok { |
| 118 | return sel, true, nil |
| 119 | } |
| 120 | var shape struct { |
| 121 | Pool string `json:"pool"` |
| 122 | } |
| 123 | if err := json.Unmarshal(r.Body, &shape); err != nil { |
| 124 | return hostclaim.ExecutorSelection{}, false, err |
| 125 | } |
| 126 | x := *a.Executors |
| 127 | x.Namespace, x.Pool = a.Namespace, shape.Pool |
| 128 | sel, err := x.Choose(ctx, obs, r) |
| 129 | return sel, err == nil, err |
| 130 | } |
| 131 | |
| 132 | func reportKey(reservation, box string) (string, error) { |
| 133 | if !safeSegment(reservation) || !safeSegment(box) { |
| 134 | return "", ErrIdentity |
| 135 | } |
| 136 | return "control/host-census/" + reservation + "/" + box + ".json", nil |
| 137 | } |
| 138 | func safeSegment(s string) bool { |
| 139 | if s == "" || s == "." || s == ".." { |
| 140 | return false |
| 141 | } |
| 142 | for _, c := range s { |
| 143 | if c == '/' || c == '\\' || c == 0 { |
| 144 | return false |
| 145 | } |
| 146 | } |
| 147 | return true |
| 148 | } |
| 149 | func CheckHostQuarantine(ctx context.Context, s release.Store, host string) error { |
| 150 | raw, _, err := s.Get(ctx, HostsQuarantineKey) |
| 151 | if errors.Is(err, os.ErrNotExist) { |
| 152 | return nil |
| 153 | } |
| 154 | if err != nil { |
| 155 | return err |
| 156 | } |
| 157 | var q HostsQuarantine |
| 158 | if decode(raw, &q) != nil || q.V != 1 || q.Hosts == nil { |
| 159 | return ErrIdentity |
| 160 | } |
| 161 | if _, ok := q.Hosts[host]; ok { |
| 162 | return ErrPending |
| 163 | } |
| 164 | return nil |
| 165 | } |
| 166 | func (b *Backfill) read(ctx context.Context) (BackfillPlan, string, error) { |
| 167 | p, tag, err := b.readPlan(ctx) |
| 168 | if err != nil { |
| 169 | return p, "", err |
| 170 | } |
| 171 | if p.Phase != "adopting" { |
| 172 | return p, "", ErrContract |
| 173 | } |
| 174 | return p, tag, nil |
| 175 | } |
| 176 | func (b *Backfill) readPlan(ctx context.Context) (BackfillPlan, string, error) { |
| 177 | var p BackfillPlan |
| 178 | raw, tag, err := b.Store.Get(ctx, BackfillKey) |
| 179 | if err != nil { |
| 180 | return p, "", err |
| 181 | } |
| 182 | if tag == "" || release.VerifyRootEnvelope(b.Directory.PublicKey, raw, &p) != nil || p.V != 1 || !safeSegment(p.ID) || p.CreatedAt.IsZero() || (p.Phase == "adopting" && (p.Hosts == nil || p.Reports == nil)) { |
| 183 | return p, "", ErrContract |
| 184 | } |
| 185 | return p, tag, nil |
| 186 | } |
| 187 | func (a ProductionAuthority) census(ctx context.Context, account, app, reservation string, expectedRevision uint64, digests map[string]string) (Census, error) { |
| 188 | out := Census{Reports: map[string]CensusReport{}} |
| 189 | roster, err := a.Directory.Read(ctx) |
| 190 | if err != nil { |
| 191 | return out, err |
| 192 | } |
| 193 | if expectedRevision != 0 && roster.Roster.Revision != expectedRevision { |
| 194 | return out, ErrPending |
| 195 | } |
| 196 | out.Roster = roster |
| 197 | apps, err := Apps(ctx, a.Store) |
| 198 | if err != nil { |
| 199 | return out, err |
| 200 | } |
| 201 | allRecords := map[string]Record{} |
| 202 | for _, p := range apps { |
| 203 | records, err := Records(ctx, a.Store, p.Account, p.Name) |
| 204 | if err != nil { |
| 205 | return out, err |
| 206 | } |
| 207 | for _, r := range records { |
| 208 | if _, ok := allRecords[r.Holder.Host]; ok { |
| 209 | return out, ErrIdentity |
| 210 | } |
| 211 | allRecords[r.Holder.Host] = r |
| 212 | if account == "" || (p.Account == account && p.Name == app) { |
| 213 | out.Records = append(out.Records, r) |
| 214 | } |
| 215 | } |
| 216 | } |
| 217 | for ref, box := range roster.Roster.Boxes { |
| 218 | // Retired membership is retained until its inventory is proven empty. |
| 219 | key, err := reportKey(reservation, ref) |
| 220 | if err != nil { |
| 221 | return out, err |
| 222 | } |
| 223 | raw, tag, err := a.Store.Get(ctx, key) |
| 224 | if err != nil { |
| 225 | return out, err |
| 226 | } |
| 227 | if digests != nil { |
| 228 | sum := sha256.Sum256(raw) |
| 229 | if digests[ref] != hex.EncodeToString(sum[:]) { |
| 230 | return out, ErrIdentity |
| 231 | } |
| 232 | } |
| 233 | var envelope CensusEnvelope |
| 234 | if decode(raw, &envelope) != nil { |
| 235 | return out, ErrIdentity |
| 236 | } |
| 237 | report := envelope.Report |
| 238 | canonical, e := json.Marshal(report) |
| 239 | if e != nil || len(box.CensusPublicKey) != ed25519.PublicKeySize || !ed25519.Verify(box.CensusPublicKey, canonical, envelope.Signature) { |
| 240 | return out, ErrIdentity |
| 241 | } |
| 242 | if tag == "" || !report.Complete || report.Box != ref || report.Enrollment != box.Enrollment || report.Reservation != reservation || report.DirectoryRevision != roster.Roster.Revision || report.BootID == "" || report.CapturedAt.IsZero() || report.Installations == nil { |
| 243 | return out, ErrIdentity |
| 244 | } |
| 245 | seen := map[string]bool{} |
| 246 | for _, i := range report.Installations { |
| 247 | if i.Host == "" || seen[i.Host] || (!i.Route && !i.Local) { |
| 248 | return out, ErrIdentity |
| 249 | } |
| 250 | seen[i.Host] = true |
| 251 | r, ok := allRecords[i.Host] |
| 252 | if !ok || i.Account != r.Holder.Account || i.App != r.Holder.App || i.Key != r.Holder.Key || i.Family != r.Holder.Family { |
| 253 | return out, ErrIdentity |
| 254 | } |
| 255 | inc, err := r.Incarnation() |
| 256 | if err != nil || (i.Incarnation != "" && i.Incarnation != inc) { |
| 257 | return out, ErrIdentity |
| 258 | } |
| 259 | } |
| 260 | out.Reports[ref] = report |
| 261 | } |
| 262 | if digests != nil && len(digests) != len(out.Reports) { |
| 263 | return out, ErrIdentity |
| 264 | } |
| 265 | fresh, err := a.Directory.Read(ctx) |
| 266 | if err != nil { |
| 267 | return out, err |
| 268 | } |
| 269 | if fresh.ETag != roster.ETag { |
| 270 | return out, ErrPending |
| 271 | } |
| 272 | return out, nil |
| 273 | } |
| 274 | func (a ProductionAuthority) Census(ctx context.Context, account, app, reservation string) (Census, error) { |
| 275 | return a.census(ctx, account, app, reservation, 0, nil) |
| 276 | } |
| 277 | func (a ProductionAuthority) Inventory(ctx context.Context, obs hostclaim.Observed) ([]string, []string, error) { |
| 278 | roster, err := a.Directory.Read(ctx) |
| 279 | if err != nil { |
| 280 | return nil, nil, err |
| 281 | } |
| 282 | if len(roster.Roster.Regions) == 0 { |
| 283 | return nil, nil, ErrContract |
| 284 | } |
| 285 | regions := make([]string, 0, len(roster.Roster.Regions)) |
| 286 | boxes := make([]string, 0, len(roster.Roster.Boxes)) |
| 287 | for r := range roster.Roster.Regions { |
| 288 | regions = append(regions, r) |
| 289 | } |
| 290 | for b := range roster.Roster.Boxes { |
| 291 | boxes = append(boxes, b) |
| 292 | } |
| 293 | sort.Strings(regions) |
| 294 | sort.Strings(boxes) |
| 295 | return regions, boxes, nil |
| 296 | } |
| 297 | func (a ProductionAuthority) Adoption(ctx context.Context, r Record) (region, box string, regions, boxes []string, err error) { |
| 298 | if a.Backfill == nil { |
| 299 | err = ErrContract |
| 300 | return |
| 301 | } |
| 302 | return a.Backfill.Adoption(ctx, r) |
| 303 | } |
| 304 | func (b *Backfill) Adoption(ctx context.Context, r Record) (region, box string, regions, boxes []string, err error) { |
| 305 | if err = CheckHostQuarantine(ctx, b.Store, r.Holder.Host); err != nil { |
| 306 | return |
| 307 | } |
| 308 | p, tag, e := b.read(ctx) |
| 309 | if e != nil { |
| 310 | err = e |
| 311 | return |
| 312 | } |
| 313 | selected, ok := p.Hosts[r.Holder.Host] |
| 314 | if !ok || selected.Key != r.Holder.Key || selected.ETag != r.ETag || selected.Acquisition == "" || selected.Region == "" || selected.Box == "" { |
| 315 | err = ErrIdentity |
| 316 | return |
| 317 | } |
| 318 | a := ProductionAuthority{Store: b.Store, Directory: b.Directory} |
| 319 | b.mu.Lock() |
| 320 | if b.cachedTag != tag { |
| 321 | b.cached = nil |
| 322 | } |
| 323 | var census Census |
| 324 | if b.cached != nil { |
| 325 | census = *b.cached |
| 326 | } else { |
| 327 | census, e = a.census(ctx, "", "", p.ID, p.DirectoryRevision, p.Reports) |
| 328 | if e == nil { |
| 329 | b.cached = &census |
| 330 | b.cachedTag = tag |
| 331 | } |
| 332 | } |
| 333 | b.mu.Unlock() |
| 334 | if e != nil { |
| 335 | err = e |
| 336 | return |
| 337 | } |
| 338 | current, e := a.Directory.Read(ctx) |
| 339 | if e != nil { |
| 340 | err = e |
| 341 | return |
| 342 | } |
| 343 | if current.ETag != census.Roster.ETag { |
| 344 | err = ErrPending |
| 345 | return |
| 346 | } |
| 347 | owners := map[string]bool{} |
| 348 | found := false |
| 349 | for _, rec := range census.Records { |
| 350 | if rec.Holder.Host == r.Holder.Host { |
| 351 | found = rec.ETag == r.ETag && rec.Holder.Key == r.Holder.Key |
| 352 | } |
| 353 | } |
| 354 | for ref, report := range census.Reports { |
| 355 | for _, i := range report.Installations { |
| 356 | if i.Host == r.Holder.Host { |
| 357 | owners[ref] = true |
| 358 | } |
| 359 | } |
| 360 | } |
| 361 | if !found || len(owners) != 1 || !owners[selected.Box] { |
| 362 | err = ErrIdentity |
| 363 | return |
| 364 | } |
| 365 | enrolled, ok := census.Roster.Roster.Boxes[selected.Box] |
| 366 | if !ok || enrolled.Region != selected.Region || enrolled.State != "active" { |
| 367 | err = ErrIdentity |
| 368 | return |
| 369 | } |
| 370 | if r.Holder.Family == "instance" { |
| 371 | var instance provision.Record |
| 372 | if json.Unmarshal(r.Body, &instance) != nil || len(instance.Replicas) != 0 || provision.ClaimEverywhere(instance) { |
| 373 | err = ErrIdentity |
| 374 | return |
| 375 | } |
| 376 | } |
| 377 | raw, _, e := b.Store.Get(ctx, domains.ClaimKey(r.Holder.Host)) |
| 378 | if e != nil { |
| 379 | err = e |
| 380 | return |
| 381 | } |
| 382 | var ownership domains.Claim |
| 383 | inc, e := r.Incarnation() |
| 384 | if e != nil { |
| 385 | err = e |
| 386 | return |
| 387 | } |
| 388 | if decode(raw, &ownership) != nil || ownership.Host != r.Holder.Host || ownership.Account != r.Holder.Account || ownership.App != r.Holder.App || ownership.Acquisition != selected.Acquisition || ownership.RecordIncarnation != inc || ownership.OwnershipState != "committed" || ownership.Pending { |
| 389 | err = ErrIdentity |
| 390 | return |
| 391 | } |
| 392 | if r.Holder.Acquisition != selected.Acquisition { |
| 393 | err = ErrIdentity |
| 394 | return |
| 395 | } |
| 396 | _, freshTag, e := b.read(ctx) |
| 397 | if e != nil { |
| 398 | err = e |
| 399 | return |
| 400 | } |
| 401 | if freshTag != tag { |
| 402 | err = ErrPending |
| 403 | return |
| 404 | } |
| 405 | for region := range census.Roster.Roster.Regions { |
| 406 | if region != selected.Region { |
| 407 | regions = append(regions, region) |
| 408 | } |
| 409 | } |
| 410 | for ref := range census.Roster.Roster.Boxes { |
| 411 | if ref != selected.Box { |
| 412 | boxes = append(boxes, ref) |
| 413 | } |
| 414 | } |
| 415 | sort.Strings(regions) |
| 416 | sort.Strings(boxes) |
| 417 | return selected.Region, selected.Box, regions, boxes, nil |
| 418 | } |
| 419 | |
amadan