| 1 | package host |
| 2 | |
| 3 | import ( |
| 4 | "bytes" |
| 5 | "context" |
| 6 | "crypto/rand" |
| 7 | "encoding/hex" |
| 8 | "encoding/json" |
| 9 | "errors" |
| 10 | "fmt" |
| 11 | "io" |
| 12 | "net" |
| 13 | "net/http" |
| 14 | "regexp" |
| 15 | "time" |
| 16 | |
| 17 | "github.com/carlosframework/platform/internal/controlname" |
| 18 | "github.com/carlosframework/platform/internal/hostclaim" |
| 19 | "github.com/carlosframework/platform/internal/moves" |
| 20 | "github.com/carlosframework/platform/internal/provision" |
| 21 | "github.com/carlosframework/platform/internal/regionname" |
| 22 | "github.com/carlosframework/platform/internal/registry" |
| 23 | "github.com/carlosframework/platform/internal/release" |
| 24 | ) |
| 25 | |
| 26 | // The held release under the host claim (moves after claim Plan 12, Task |
| 27 | // 7b; task-7b-design.md with platform-1e's binding review of 2026-10-12). |
| 28 | // |
| 29 | // Under the claim the sandboxed agent cannot write /opt/carlos, so it can |
| 30 | // neither stage T7's release nor publish live/<host> for the held row. It |
| 31 | // asks root instead, over the root instance-control socket (claim Plan 11's |
| 32 | // R10 transport: POST /claim/held-release, the peer authenticated as the |
| 33 | // edge unit's main process), never through a file root would read by name |
| 34 | // (#267). There are two actions: |
| 35 | // |
| 36 | // - stage: T7. Root stages and verifies the pinned release into the |
| 37 | // shared version store (release.StageHeldVerified) and never touches |
| 38 | // live/<host>. The agent then verifies the staged bytes itself; root's |
| 39 | // answer is a hint. |
| 40 | // - publish: the landing's configured step, a dirty return's configured |
| 41 | // step and a probe's release. Root publishes live/<host> inside its own |
| 42 | // held-release selection (release.MaterializeHeldUnder with the |
| 43 | // held-stage lock), and writes a receipt keyed by the session and the |
| 44 | // render hash, so a replay after success is a no-op with the same |
| 45 | // answer. |
| 46 | // |
| 47 | // Every request field is an expectation root compares with authoritative |
| 48 | // state it reads itself (the claim, the record, the registry row, the |
| 49 | // render journal); none grants anything. The route is new, so an older root |
| 50 | // answers 404 and never reads a stage as a publication, and the old file |
| 51 | // relay (release.ServeHeldStageRequests) is never handed a request by this |
| 52 | // path. |
| 53 | |
| 54 | const ( |
| 55 | heldReleaseSchema = "carlos.moves.held-release/1" |
| 56 | heldReleaseStage = "stage" |
| 57 | heldReleasePublish = "publish" |
| 58 | heldReleaseOutcomeStaged = "staged" |
| 59 | heldReleaseOutcomePublished = "published" |
| 60 | heldReleasePath = "/claim/held-release" |
| 61 | // heldReleaseRequestMax bounds a request body: its fields fit many times |
| 62 | // over. |
| 63 | heldReleaseRequestMax = 4096 |
| 64 | // heldReleaseDeadline bounds one call. A stage may download the build, |
| 65 | // as T7 does in the agent with the claim off. |
| 66 | heldReleaseDeadline = 15 * time.Minute |
| 67 | // heldReleaseConcurrency bounds the held-release requests root serves at |
| 68 | // once across hosts (fix round 2, P3-4); each host has one at a time. |
| 69 | heldReleaseConcurrency = 3 |
| 70 | ) |
| 71 | |
| 72 | var ( |
| 73 | errHeldReleaseShape = errors.New("moves: held release: the request is malformed") |
| 74 | errHeldReleaseUnavailable = errors.New("moves: held release: root serves no held-release requests") |
| 75 | // errHeldReleaseNotYet is root finding no render hash journalled for |
| 76 | // the publication (platform-1e Q2): it refuses at once, holding nothing, |
| 77 | // and the agent asks again on a later pass. |
| 78 | errHeldReleaseNotYet = errors.New("moves: held release: the held render's hash is not journalled yet") |
| 79 | errHeldReleaseUnsafe = errors.New("moves: held release: the render journal is not a regular file with one link and a bounded size") |
| 80 | // errHeldReleaseReceipt is a publication root made whose receipt did not |
| 81 | // persist (fix round 1, item 1): root never answers published without the |
| 82 | // evidence a replay is answered from, so the agent asks again, and the |
| 83 | // retry runs the whole path (the same render, the same selection checks) |
| 84 | // and publishes the same release. |
| 85 | errHeldReleaseReceipt = errors.New("moves: held release: published, but root's receipt did not persist") |
| 86 | // errHeldReleaseBusy is root at its bound (fix round 2, P3-4): a request |
| 87 | // for the host is already in flight, or heldReleaseConcurrency are. It |
| 88 | // answers 503, which the agent reads as unavailable and retries. |
| 89 | errHeldReleaseBusy = errors.New("moves: held release: root is busy with another request") |
| 90 | ) |
| 91 | |
| 92 | // heldReleaseRequest is one request. Kind, Phase, Incarnation, Stamp, |
| 93 | // Session and Generation are the claim tuple the agent observed; MoveID is |
| 94 | // the claim's op id; HoldOp the op its registry row is held by (the |
| 95 | // rollback id for a return); RenderHash, on a publication, the held |
| 96 | // render's journalled hash ("sha256:<hex>"). |
| 97 | type heldReleaseRequest struct { |
| 98 | Schema string `json:"schema"` |
| 99 | RequestID string `json:"request_id"` |
| 100 | Action string `json:"action"` |
| 101 | Host string `json:"host"` |
| 102 | Account string `json:"account"` |
| 103 | App string `json:"app"` |
| 104 | Version string `json:"version"` |
| 105 | MoveID string `json:"move_id"` |
| 106 | HoldOp string `json:"hold_op"` |
| 107 | Box string `json:"box"` |
| 108 | Region string `json:"region"` |
| 109 | Incarnation string `json:"incarnation"` |
| 110 | Stamp uint64 `json:"stamp"` |
| 111 | Session uint64 `json:"session"` |
| 112 | Generation string `json:"generation"` |
| 113 | Kind string `json:"kind"` |
| 114 | Phase string `json:"phase"` |
| 115 | RenderHash string `json:"render_hash,omitempty"` |
| 116 | } |
| 117 | |
| 118 | type heldReleaseResponse struct { |
| 119 | Schema string `json:"schema"` |
| 120 | RequestID string `json:"request_id"` |
| 121 | Outcome string `json:"outcome"` |
| 122 | Replayed bool `json:"replayed,omitempty"` |
| 123 | } |
| 124 | |
| 125 | var ( |
| 126 | heldReleaseIDShape = regexp.MustCompile(`^[A-Za-z0-9_-]{1,64}$`) |
| 127 | heldReleaseBoxShape = regexp.MustCompile(`^[A-Za-z0-9._-]{1,128}$`) |
| 128 | heldReleaseHexShape = regexp.MustCompile(`^[0-9a-f]{64}$`) |
| 129 | heldReleaseRenderShape = regexp.MustCompile(`^sha256:[0-9a-f]{64}$`) |
| 130 | heldReleaseTokenShape = regexp.MustCompile(`^[A-Za-z0-9._:@-]{0,128}$`) |
| 131 | ) |
| 132 | |
| 133 | // validate is the shape root trusts before any read: every name a path |
| 134 | // component or a key could be built from, and the action's own fields. |
| 135 | func (q heldReleaseRequest) validate() error { |
| 136 | bad := func(what string) error { return fmt.Errorf("%w: %s", errHeldReleaseShape, what) } |
| 137 | switch { |
| 138 | case q.Schema != heldReleaseSchema: |
| 139 | return bad("schema") |
| 140 | case q.Action != heldReleaseStage && q.Action != heldReleasePublish: |
| 141 | return bad("action") |
| 142 | case !heldReleaseIDShape.MatchString(q.RequestID): |
| 143 | return bad("request id") |
| 144 | case !controlname.ValidHostShape(q.Host): |
| 145 | return bad("host") |
| 146 | case !release.ValidAccount(q.Account): |
| 147 | return bad("account") |
| 148 | case !release.ValidApp(q.App): |
| 149 | return bad("app") |
| 150 | case !release.ValidVersion(q.Version): |
| 151 | return bad("version") |
| 152 | case !moves.ValidOpID(q.MoveID) || moves.OpIDKind(q.MoveID) != "move": |
| 153 | return bad("move id") |
| 154 | case !moves.ValidOpID(q.HoldOp): |
| 155 | return bad("hold op") |
| 156 | case !heldReleaseBoxShape.MatchString(q.Box): |
| 157 | return bad("box") |
| 158 | case q.Region != regionname.HomeRegion && (q.Region == "" || regionname.PlacementRegion(q.Region) != q.Region): |
| 159 | return bad("region") |
| 160 | case !heldReleaseHexShape.MatchString(q.Incarnation): |
| 161 | return bad("incarnation") |
| 162 | case q.Stamp == 0: |
| 163 | return bad("stamp") |
| 164 | case !heldReleaseTokenShape.MatchString(q.Generation) || len(q.Phase) > 64 || !heldReleaseTokenShape.MatchString(q.Phase): |
| 165 | return bad("generation or phase") |
| 166 | } |
| 167 | switch q.Kind { |
| 168 | case hostclaim.KindMoving, hostclaim.KindLanding, hostclaim.KindRestoring, hostclaim.KindReverting: |
| 169 | default: |
| 170 | return bad("kind") |
| 171 | } |
| 172 | if q.Action == heldReleasePublish && !heldReleaseRenderShape.MatchString(q.RenderHash) { |
| 173 | return bad("render hash") |
| 174 | } |
| 175 | if q.Action == heldReleaseStage && q.RenderHash != "" { |
| 176 | return bad("a stage names no render") |
| 177 | } |
| 178 | return nil |
| 179 | } |
| 180 | |
| 181 | // decodeHeldReleaseRequest reads one request body: bounded, one JSON object, |
| 182 | // no field the protocol does not carry. |
| 183 | func decodeHeldReleaseRequest(r io.Reader) (heldReleaseRequest, error) { |
| 184 | body, err := io.ReadAll(io.LimitReader(r, heldReleaseRequestMax+1)) |
| 185 | if err != nil || len(body) > heldReleaseRequestMax { |
| 186 | return heldReleaseRequest{}, fmt.Errorf("%w: unreadable or oversized", errHeldReleaseShape) |
| 187 | } |
| 188 | dec := json.NewDecoder(bytes.NewReader(body)) |
| 189 | dec.DisallowUnknownFields() |
| 190 | var q heldReleaseRequest |
| 191 | if err := dec.Decode(&q); err != nil { |
| 192 | return heldReleaseRequest{}, fmt.Errorf("%w: %v", errHeldReleaseShape, err) |
| 193 | } |
| 194 | if err := dec.Decode(&struct{}{}); !errors.Is(err, io.EOF) { |
| 195 | return heldReleaseRequest{}, fmt.Errorf("%w: trailing data", errHeldReleaseShape) |
| 196 | } |
| 197 | return q, nil |
| 198 | } |
| 199 | |
| 200 | func newHeldReleaseRequestID() string { |
| 201 | var b [16]byte |
| 202 | if _, err := rand.Read(b[:]); err != nil { |
| 203 | return fmt.Sprintf("t%d", time.Now().UnixNano()) |
| 204 | } |
| 205 | return hex.EncodeToString(b[:]) |
| 206 | } |
| 207 | |
| 208 | // heldReleaseRelay is the agent's way to root. |
| 209 | type heldReleaseRelay interface { |
| 210 | HeldRelease(context.Context, heldReleaseRequest) (heldReleaseResponse, error) |
| 211 | } |
| 212 | |
| 213 | // heldReleaseRootClient is the agent's client of POST /claim/held-release |
| 214 | // on the root instance-control socket. |
| 215 | type heldReleaseRootClient struct { |
| 216 | Socket string |
| 217 | // Dial is the dialer; nil dials Socket as a unix socket. |
| 218 | Dial func(context.Context, string) (net.Conn, error) |
| 219 | } |
| 220 | |
| 221 | func newHeldReleaseRootClient(socket string) *heldReleaseRootClient { |
| 222 | return &heldReleaseRootClient{Socket: socket} |
| 223 | } |
| 224 | |
| 225 | // HeldRelease sends q and returns root's answer for exactly q. Root's |
| 226 | // refusals come back typed: a stale or forged request wraps |
| 227 | // release.ErrStaleHeldStage, a release D3 does not admit |
| 228 | // release.ErrHeldNotProven, no render journal errHeldReleaseNotYet, and a |
| 229 | // root with no such route (an older build, or one without its providers) |
| 230 | // errHeldReleaseUnavailable. |
| 231 | func (c *heldReleaseRootClient) HeldRelease(ctx context.Context, q heldReleaseRequest) (heldReleaseResponse, error) { |
| 232 | if c == nil || c.Socket == "" { |
| 233 | return heldReleaseResponse{}, errHeldReleaseUnavailable |
| 234 | } |
| 235 | body, err := json.Marshal(q) |
| 236 | if err != nil { |
| 237 | return heldReleaseResponse{}, err |
| 238 | } |
| 239 | ctx, cancel := context.WithTimeout(ctx, heldReleaseDeadline) |
| 240 | defer cancel() |
| 241 | dial := c.Dial |
| 242 | if dial == nil { |
| 243 | dial = func(ctx context.Context, path string) (net.Conn, error) { |
| 244 | return (&net.Dialer{}).DialContext(ctx, "unix", path) |
| 245 | } |
| 246 | } |
| 247 | transport := &http.Transport{DisableKeepAlives: true, DialContext: func(ctx context.Context, _, _ string) (net.Conn, error) { return dial(ctx, c.Socket) }} |
| 248 | defer transport.CloseIdleConnections() |
| 249 | req, err := http.NewRequestWithContext(ctx, http.MethodPost, "http://control"+heldReleasePath, bytes.NewReader(body)) |
| 250 | if err != nil { |
| 251 | return heldReleaseResponse{}, err |
| 252 | } |
| 253 | req.Header.Set("Content-Type", "application/json") |
| 254 | resp, err := (&http.Client{Transport: transport}).Do(req) |
| 255 | if err != nil { |
| 256 | return heldReleaseResponse{}, fmt.Errorf("moves: held release: %w", err) |
| 257 | } |
| 258 | defer resp.Body.Close() |
| 259 | defer io.Copy(io.Discard, io.LimitReader(resp.Body, 4<<10)) |
| 260 | switch resp.StatusCode { |
| 261 | case http.StatusOK: |
| 262 | case http.StatusBadRequest: |
| 263 | return heldReleaseResponse{}, errHeldReleaseShape |
| 264 | case http.StatusForbidden: |
| 265 | return heldReleaseResponse{}, ErrClaimStopIdentity |
| 266 | case http.StatusConflict: |
| 267 | return heldReleaseResponse{}, staleHeldStage("root refused the held release") |
| 268 | case http.StatusUnprocessableEntity: |
| 269 | return heldReleaseResponse{}, fmt.Errorf("%w: root's proof gate refused it", release.ErrHeldNotProven) |
| 270 | case http.StatusTooEarly: |
| 271 | return heldReleaseResponse{}, errHeldReleaseNotYet |
| 272 | case http.StatusNotFound, http.StatusMethodNotAllowed, http.StatusServiceUnavailable: |
| 273 | return heldReleaseResponse{}, errHeldReleaseUnavailable |
| 274 | default: |
| 275 | return heldReleaseResponse{}, fmt.Errorf("moves: held release: root answered %d", resp.StatusCode) |
| 276 | } |
| 277 | dec := json.NewDecoder(io.LimitReader(resp.Body, heldReleaseRequestMax)) |
| 278 | dec.DisallowUnknownFields() |
| 279 | var out heldReleaseResponse |
| 280 | if err := dec.Decode(&out); err != nil { |
| 281 | return heldReleaseResponse{}, fmt.Errorf("moves: held release: root's answer: %w", err) |
| 282 | } |
| 283 | want := heldReleaseOutcomeStaged |
| 284 | if q.Action == heldReleasePublish { |
| 285 | want = heldReleaseOutcomePublished |
| 286 | } |
| 287 | if out.Schema != heldReleaseSchema || out.RequestID != q.RequestID || out.Outcome != want { |
| 288 | return heldReleaseResponse{}, fmt.Errorf("moves: held release: root's answer is not for this request: %+v", out) |
| 289 | } |
| 290 | return out, nil |
| 291 | } |
| 292 | |
| 293 | // heldReleaseVia is how the agent's held-release hooks reach root: the |
| 294 | // relay (nil: no root path, Task 7's local behaviour), this box's |
| 295 | // placement code, sent as its engine region, and the claim reader (nil: |
| 296 | // hostclaim.Read over the hook's store; a test seam). |
| 297 | type heldReleaseVia struct { |
| 298 | Relay heldReleaseRelay |
| 299 | Region string |
| 300 | Claims func(context.Context, string) (hostclaim.Observed, error) |
| 301 | } |
| 302 | |
| 303 | func (v heldReleaseVia) claim(ctx context.Context, s release.Store, host string) (hostclaim.Observed, error) { |
| 304 | if v.Claims != nil { |
| 305 | return v.Claims(ctx, host) |
| 306 | } |
| 307 | return hostclaim.Read(ctx, s, host) |
| 308 | } |
| 309 | |
| 310 | // request builds the agent's request for action from its own reads: the |
| 311 | // claim tuple as it observes it now, moveID ("" takes the claim's op id) |
| 312 | // and holdOp from its record or route, and on a publication the render hash |
| 313 | // configured journalled under holdOp. None of it is authority: root reads |
| 314 | // each again and compares. |
| 315 | func (v heldReleaseVia) request(ctx context.Context, s release.Store, action, host, account, app, version, moveID, holdOp, box, dataDir string) (heldReleaseRequest, error) { |
| 316 | obs, err := v.claim(ctx, s, host) |
| 317 | if err != nil { |
| 318 | return heldReleaseRequest{}, err |
| 319 | } |
| 320 | if obs.Absent || obs.Claim == nil || obs.Claim.Holder == nil { |
| 321 | return heldReleaseRequest{}, staleHeldStage("the host has no claim") |
| 322 | } |
| 323 | c := obs.Claim |
| 324 | if moveID == "" { |
| 325 | moveID = c.Op.ID |
| 326 | } |
| 327 | q := heldReleaseRequest{Schema: heldReleaseSchema, RequestID: newHeldReleaseRequestID(), Action: action, Host: host, Account: account, App: app, |
| 328 | Version: version, MoveID: moveID, HoldOp: holdOp, Box: box, Region: regionname.PlacementRegion(v.Region), |
| 329 | Incarnation: c.Holder.Incarnation, Stamp: c.Holder.Stamp, Session: c.Holder.Session, Generation: c.Holder.Generation, |
| 330 | Kind: c.Op.Kind, Phase: c.Op.Phase} |
| 331 | if action == heldReleasePublish { |
| 332 | j, err := moves.ReadJournal(dataDir, host, holdOp) |
| 333 | if err != nil { |
| 334 | return heldReleaseRequest{}, err |
| 335 | } |
| 336 | if q.RenderHash = j.Steps[landStepRendered]; q.RenderHash == "" { |
| 337 | return heldReleaseRequest{}, errHeldReleaseNotYet |
| 338 | } |
| 339 | } |
| 340 | return q, nil |
| 341 | } |
| 342 | |
| 343 | // heldReleaseHoldOp is the op rec's registry row is held by: the rollback |
| 344 | // id while a rollback runs (ruling J3), else the move id. |
| 345 | func heldReleaseHoldOp(rec provision.Record) string { |
| 346 | if m := rec.Move; m != nil && m.State == provision.MoveRollingBack && m.Rollback != nil && m.Rollback.ID != "" { |
| 347 | return m.Rollback.ID |
| 348 | } |
| 349 | if rec.Move == nil { |
| 350 | return "" |
| 351 | } |
| 352 | return rec.Move.ID |
| 353 | } |
| 354 | |
| 355 | // stage asks root to stage T7's release for rec (the move's target, its |
| 356 | // row held, if at all, by the move id). |
| 357 | func (v heldReleaseVia) stage(ctx context.Context, s release.Store, rec provision.Record, box, version string) error { |
| 358 | q, err := v.request(ctx, s, heldReleaseStage, rec.Host, rec.Account, rec.App, version, rec.Move.ID, rec.Move.ID, box, "") |
| 359 | if err != nil { |
| 360 | return err |
| 361 | } |
| 362 | _, err = v.Relay.HeldRelease(ctx, q) |
| 363 | return err |
| 364 | } |
| 365 | |
| 366 | // publish asks root to publish version for rt under holdOp's render |
| 367 | // journal. |
| 368 | func (v heldReleaseVia) publish(ctx context.Context, s release.Store, rt registry.Route, moveID, holdOp, box, dataDir, version string) error { |
| 369 | q, err := v.request(ctx, s, heldReleasePublish, rt.Host, release.RouteAccount(rt), rt.App, version, moveID, holdOp, box, dataDir) |
| 370 | if err != nil { |
| 371 | return err |
| 372 | } |
| 373 | _, err = v.Relay.HeldRelease(ctx, q) |
| 374 | return err |
| 375 | } |
| 376 | |
amadan