| 1 | package hostops |
| 2 | |
| 3 | import ( |
| 4 | "bytes" |
| 5 | "context" |
| 6 | "crypto/ed25519" |
| 7 | "encoding/json" |
| 8 | "errors" |
| 9 | "io" |
| 10 | "math" |
| 11 | "os" |
| 12 | "reflect" |
| 13 | "strings" |
| 14 | "time" |
| 15 | |
| 16 | "github.com/carlosframework/platform/internal/hostclaim" |
| 17 | "github.com/carlosframework/platform/internal/release" |
| 18 | ) |
| 19 | |
| 20 | // Lifecycle claims (control/hosts/<host>.json) are the claim engine's CAS |
| 21 | // root. The built Tasks 2-6 engine writes them through the data handle, so |
| 22 | // the data boundary cannot refuse them outright. It admits a claim write |
| 23 | // only under a lifecycle grant carried in ctx, and validates the exact |
| 24 | // old->new claim against the grant's scope. A grant is minted only by a |
| 25 | // writer-issued handle (its own app/operation) or by RecoveryPass (stale |
| 26 | // operations and frozen-backfill adoption only). A handler holding |
| 27 | // Config.Releases and no grant cannot write any claim. |
| 28 | |
| 29 | type lifecycleGrantKey struct{} |
| 30 | |
| 31 | type lifecycleGrant struct { |
| 32 | kind string // "app" or "recovery" |
| 33 | account, app, appID string |
| 34 | operation string |
| 35 | // exclusive: the grant is an OperationHandle's (its op must still hold |
| 36 | // admission); otherwise a StartHandle's (its registration must be live |
| 37 | // at the handle's epoch with the handle's owner). |
| 38 | exclusive bool |
| 39 | owner RegistrationOwner |
| 40 | epoch uint64 |
| 41 | w *ControlWriter |
| 42 | // registration and incarnation are a StartHandle's registration kind |
| 43 | // and saved holder incarnation. A reset-wake-retry grant admits only |
| 44 | // the exact ResetWakeRetry transition for that incarnation (review R2). |
| 45 | registration, incarnation string |
| 46 | } |
| 47 | |
| 48 | // Lifecycle returns ctx carrying this registration's claim grant. The grant |
| 49 | // is checked against admission on every claim write: it dies with the |
| 50 | // registration. |
| 51 | func (h *StartHandle) Lifecycle(ctx context.Context) context.Context { |
| 52 | if h.kind == RegistrationIngest { |
| 53 | return ctx |
| 54 | } |
| 55 | return withLifecycleGrant(ctx, lifecycleGrant{kind: "app", account: h.account, app: h.app, appID: h.appID, operation: h.id, owner: h.owner, epoch: h.epoch, w: h.w, registration: h.kind, incarnation: h.incarnation}) |
| 56 | } |
| 57 | |
| 58 | // Lifecycle returns ctx carrying this operation's claim grant. The grant is |
| 59 | // checked against admission on every claim write: it dies with the op. |
| 60 | func (h *OperationHandle) Lifecycle(ctx context.Context) context.Context { |
| 61 | return withLifecycleGrant(ctx, lifecycleGrant{kind: "app", account: h.account, app: h.app, appID: h.appID, operation: h.id, exclusive: true, owner: h.owner, w: h.w}) |
| 62 | } |
| 63 | |
| 64 | // grantLive re-reads admission: an app grant authorizes nothing once its |
| 65 | // registration lapsed or was drained, or once its operation ended. |
| 66 | func grantLive(ctx context.Context, raw release.Store, g lifecycleGrant) (bool, error) { |
| 67 | a, err := ReadAdmission(ctx, raw, g.account, g.app, g.appID) |
| 68 | if errors.Is(err, os.ErrNotExist) { |
| 69 | return false, nil |
| 70 | } |
| 71 | if err != nil { |
| 72 | return false, err |
| 73 | } |
| 74 | v := a.Value |
| 75 | if g.exclusive { |
| 76 | return v.State == "closed" && v.Op != nil && v.Op.ID == g.operation && v.Op.AppID == g.appID, nil |
| 77 | } |
| 78 | r, ok := v.Registrations[g.operation] |
| 79 | return ok && r.Owner == g.owner && r.AppID == g.appID && r.Epoch == g.epoch && v.Epoch == g.epoch && RegistrationLive(r, g.w.now()), nil |
| 80 | } |
| 81 | |
| 82 | func withLifecycleGrant(ctx context.Context, g lifecycleGrant) context.Context { |
| 83 | inherited, _ := ctx.Value(lifecycleGrantKey{}).([]lifecycleGrant) |
| 84 | return context.WithValue(ctx, lifecycleGrantKey{}, append(append([]lifecycleGrant(nil), inherited...), g)) |
| 85 | } |
| 86 | |
| 87 | // RecoveryPass runs the built recovery pass with a recovery grant. That |
| 88 | // grant only lets a stale operation advance or clear (never install a new |
| 89 | // holder or a new operation) and lets frozen backfill adopt an absent |
| 90 | // claim for an existing matching record. |
| 91 | func (w *ControlWriter) RecoveryPass(ctx context.Context, r Recovery) error { |
| 92 | return w.RunRecovery(ctx, r.Pass) |
| 93 | } |
| 94 | |
| 95 | // RunRecovery runs one console recovery body (frozen-backfill reconcile and |
| 96 | // the recovery pass) under the recovery grant. The grant is constrained by |
| 97 | // recoveryGrantAllows, so a caller gains nothing beyond what recovery itself |
| 98 | // may do: advance or clear a stale operation, or adopt for a signed plan. |
| 99 | func (w *ControlWriter) RunRecovery(ctx context.Context, body func(context.Context) error) error { |
| 100 | if err := w.ready(); err != nil { |
| 101 | return err |
| 102 | } |
| 103 | if body == nil { |
| 104 | return ErrContract |
| 105 | } |
| 106 | scope := &moveRecoveryPassScope{} |
| 107 | scope.live.Store(true) |
| 108 | defer scope.live.Store(false) |
| 109 | ctx = context.WithValue(ctx, moveRecoveryPassKey{}, scope) |
| 110 | return body(withLifecycleGrant(ctx, lifecycleGrant{kind: "recovery", w: w})) |
| 111 | } |
| 112 | |
| 113 | // decodeClaim is strict (unknown fields and trailing data refuse) and |
| 114 | // validates the decoded claim, as hostclaim.Read does. |
| 115 | func decodeClaim(b []byte) (*hostclaim.Claim, error) { |
| 116 | if len(b) == 0 { |
| 117 | return nil, nil |
| 118 | } |
| 119 | dec := json.NewDecoder(bytes.NewReader(b)) |
| 120 | dec.DisallowUnknownFields() |
| 121 | var c hostclaim.Claim |
| 122 | if err := dec.Decode(&c); err != nil { |
| 123 | return nil, err |
| 124 | } |
| 125 | if _, err := dec.Token(); err != io.EOF { |
| 126 | return nil, ErrIdentity |
| 127 | } |
| 128 | if err := c.Validate(); err != nil { |
| 129 | return nil, err |
| 130 | } |
| 131 | return &c, nil |
| 132 | } |
| 133 | |
| 134 | // claimTransitionShape is review P1-1's per-write validation: same host as |
| 135 | // the key, generation exactly old+1 (1 from absence), same schema version. |
| 136 | func claimTransitionShape(host string, old, next *hostclaim.Claim) bool { |
| 137 | if next.Host != host { |
| 138 | return false |
| 139 | } |
| 140 | if old == nil { |
| 141 | return next.Gen == 1 |
| 142 | } |
| 143 | return old.Host == host && next.V == old.V && old.Gen < math.MaxUint64 && next.Gen == old.Gen+1 |
| 144 | } |
| 145 | |
| 146 | func holderOf(c *hostclaim.Claim) *hostclaim.Holder { |
| 147 | if c == nil { |
| 148 | return nil |
| 149 | } |
| 150 | return c.Holder |
| 151 | } |
| 152 | func opOf(c *hostclaim.Claim) hostclaim.Op { |
| 153 | if c == nil { |
| 154 | return hostclaim.Op{} |
| 155 | } |
| 156 | return c.Op |
| 157 | } |
| 158 | |
| 159 | // authorizeClaimWrite is the classifier's lifecycle-claim rule. |
| 160 | func authorizeClaimWrite(ctx context.Context, raw release.Store, host string, t MutationTarget) error { |
| 161 | grants, _ := ctx.Value(lifecycleGrantKey{}).([]lifecycleGrant) |
| 162 | if len(grants) == 0 { |
| 163 | return ErrPending |
| 164 | } |
| 165 | old, err := decodeClaim(t.Before) |
| 166 | if err != nil { |
| 167 | return ErrIdentity |
| 168 | } |
| 169 | next, err := decodeClaim(t.After) |
| 170 | if err != nil || next == nil { |
| 171 | return ErrIdentity |
| 172 | } |
| 173 | if !claimTransitionShape(host, old, next) || |
| 174 | (old == nil && next.Origins != nil) || |
| 175 | (old != nil && !reflect.DeepEqual(old.Origins, next.Origins)) { |
| 176 | return ErrIdentity |
| 177 | } |
| 178 | for _, g := range grants { |
| 179 | if g.w == nil || !operationHomeMatches(g.w.homeRegion, old) || !operationHomeMatches(g.w.homeRegion, next) { |
| 180 | continue |
| 181 | } |
| 182 | switch g.kind { |
| 183 | case "app": |
| 184 | if !appGrantAllows(g, old, next) || g.registration == RegistrationWakeRetryReset && !resetWakeRetryTransition(g.incarnation, old, next) { |
| 185 | continue |
| 186 | } |
| 187 | if live, err := grantLive(ctx, raw, g); err != nil { |
| 188 | return err |
| 189 | } else if live { |
| 190 | return nil |
| 191 | } |
| 192 | case "recovery": |
| 193 | if ok, err := recoveryGrantAllows(ctx, raw, g.w, host, old, next); err != nil { |
| 194 | return err |
| 195 | } else if ok { |
| 196 | return nil |
| 197 | } |
| 198 | } |
| 199 | } |
| 200 | return ErrPending |
| 201 | } |
| 202 | |
| 203 | func sameApp(h *hostclaim.Holder, g lifecycleGrant) bool { |
| 204 | return h != nil && h.Account == g.account && h.App == g.app |
| 205 | } |
| 206 | |
| 207 | func appGrantAllows(g lifecycleGrant, old, next *hostclaim.Claim) bool { |
| 208 | oh, nh := holderOf(old), holderOf(next) |
| 209 | // Installing (or replacing) a holder: only this grant's own app. |
| 210 | if nh != nil && (oh == nil || oh.Incarnation != nh.Incarnation) && !sameApp(nh, g) { |
| 211 | return false |
| 212 | } |
| 213 | if oh != nil && !sameApp(oh, g) { |
| 214 | // A foreign holder is never replaced or removed by an app grant |
| 215 | // (v2 let a write that merely NAMED its own op ID in next replace |
| 216 | // it), and only an op this grant already owns on that claim may be |
| 217 | // advanced or cleared. Installing this grant's op over a foreign |
| 218 | // holder is refused. |
| 219 | if nh == nil || nh.Account != oh.Account || nh.App != oh.App || nh.Incarnation != oh.Incarnation { |
| 220 | return false |
| 221 | } |
| 222 | return opOf(old).ID == g.operation && (opOf(next).ID == g.operation || opOf(next).ID == "") |
| 223 | } |
| 224 | if oh == nil { |
| 225 | // No holder: never overwrite another operation's op (a foreign |
| 226 | // purge fence or creating op); stale ops are recovery's. |
| 227 | if id := opOf(old).ID; id != "" && id != g.operation { |
| 228 | return false |
| 229 | } |
| 230 | } |
| 231 | return sameApp(oh, g) || sameApp(nh, g) || opOf(old).ID == g.operation || opOf(next).ID == g.operation |
| 232 | } |
| 233 | |
| 234 | // resetWakeRetryTransition reports whether old->next is exactly what |
| 235 | // hostclaim.ResetWakeRetry writes for the registered holder incarnation: |
| 236 | // the same claim with gen advanced and the holder's wake-retry episode |
| 237 | // replaced by a fresh open episode that carries the operator request. |
| 238 | func resetWakeRetryTransition(incarnation string, old, next *hostclaim.Claim) bool { |
| 239 | oh, nh := holderOf(old), holderOf(next) |
| 240 | if incarnation == "" || oh == nil || nh == nil || oh.Incarnation != incarnation || oh.WakeRetry == nil || nh.WakeRetry == nil { |
| 241 | return false |
| 242 | } |
| 243 | r := nh.WakeRetry |
| 244 | if r.EpisodeID == "" || r.EpisodeID == oh.WakeRetry.EpisodeID || r.State != "open" || r.ResetRequest == "" || |
| 245 | r.AttemptsReserved != 0 || !r.NotBefore.IsZero() || r.Attempt != nil { |
| 246 | return false |
| 247 | } |
| 248 | want, holder := *old, *oh |
| 249 | holder.WakeRetry = r |
| 250 | want.Holder, want.Gen = &holder, next.Gen |
| 251 | a, err := json.Marshal(&want) |
| 252 | if err != nil { |
| 253 | return false |
| 254 | } |
| 255 | b, err := json.Marshal(next) |
| 256 | return err == nil && bytes.Equal(a, b) |
| 257 | } |
| 258 | |
| 259 | func recoveryGrantAllows(ctx context.Context, raw release.Store, w *ControlWriter, host string, old, next *hostclaim.Claim) (bool, error) { |
| 260 | oh, nh := holderOf(old), holderOf(next) |
| 261 | if old != nil { |
| 262 | if moveRecoveryKind(old.Op.Kind) { |
| 263 | return false, nil |
| 264 | } |
| 265 | op := old.Op |
| 266 | stale := op.Kind == hostclaim.KindParking && len(old.Obligations) == 0 || |
| 267 | op.Kind != "" && op.Kind != hostclaim.KindServing && !op.StartedAt.IsZero() && w.now().After(op.StartedAt.Add(5*time.Minute)) |
| 268 | if stale && (nh == nil || (oh != nil && oh.Incarnation == nh.Incarnation)) && (next.Op.ID == op.ID || next.Op.ID == "") { |
| 269 | return true, nil |
| 270 | } |
| 271 | } |
| 272 | // Frozen-backfill adoption: an absent or free claim gains the holder of an |
| 273 | // existing matching record, only while a signed plan is adopting. |
| 274 | if nh == nil || (old != nil && (oh != nil || old.Op.Kind != "")) { |
| 275 | return false, nil |
| 276 | } |
| 277 | rawPlan, _, err := raw.Get(ctx, BackfillKey) |
| 278 | if errors.Is(err, os.ErrNotExist) { |
| 279 | return false, nil |
| 280 | } |
| 281 | if err != nil { |
| 282 | return false, err |
| 283 | } |
| 284 | var plan BackfillPlan |
| 285 | if release.VerifyRootEnvelope(w.pub, rawPlan, &plan) != nil || plan.Phase != "adopting" { |
| 286 | return false, nil |
| 287 | } |
| 288 | if !strings.HasPrefix(nh.Key, "apps/"+nh.Account+"/"+nh.App+"/") { |
| 289 | return false, ErrIdentity |
| 290 | } |
| 291 | r, err := ReadRecord(ctx, raw, nh.Key) |
| 292 | if errors.Is(err, os.ErrNotExist) { |
| 293 | return false, nil |
| 294 | } |
| 295 | if err != nil { |
| 296 | return false, err |
| 297 | } |
| 298 | return r.Holder.Host == host && matches(r, nh), nil |
| 299 | } |
| 300 | |
| 301 | // authorizeSignedControl admits a root-signed control object (the roster, |
| 302 | // the backfill plan) only with a valid root envelope. |
| 303 | func authorizeSignedControl(pub ed25519.PublicKey, t MutationTarget) error { |
| 304 | if t.Method != "Put" { |
| 305 | return ErrIdentity |
| 306 | } |
| 307 | var v json.RawMessage |
| 308 | if release.VerifyRootEnvelope(pub, t.After, &v) != nil { |
| 309 | return ErrIdentity |
| 310 | } |
| 311 | return nil |
| 312 | } |
| 313 | |
amadan