fix(analytics): scope the dimension-label lookup to the referenced object's RLS (#3602 residual 1) (#3639)

os-zhuang · claude · web-flow · commit 415254c32d40 · 2026-07-27T21:05:07.000+08:00
* fix(analytics): scope dimension-label lookup to the referenced object's RLS (#3602) A dataset that groups by a lookup/master_detail dimension resolves the grouped FK ids to the related record's display name via a per-record read (`group by id`) dressed as an aggregate. That read carried no read scope, so it revealed related-record display names whenever the referenced object's RLS is stricter than the base object whose rows carry the id — a user could read a name the referenced object's own RLS would hide. (Same-object and looser-referenced cases were already safe: the ids come from the post-#3597 scoped aggregate. This closes the stricter-referenced case, the one residual flagged in #3602.) The label lookup now applies the REFERENCED object's own read scope, bound to the request via the same getReadScope provider the aggregate path uses, composed with `$and` (never key-merge) so the id predicate can't displace it. Fail-closed: if that object's scope can't be resolved, the dimension's labels are skipped (raw id renders) instead of fetched unscoped. No change when no read-scope provider is configured. `DimensionLabelDeps.fetchRecordLabels` gains an optional `scope` arg and `resolveDimensionLabels` an optional `resolveScope` resolver — both service-analytics-internal, no spec/contract change. Tests: dimension-labels unit cases prove the referenced-object scope is threaded and that resolution failure fails closed; a query-dataset integration case proves AnalyticsService.queryDataset binds the referenced object's scope (not the base object's) — reverting the wiring turns it red. Co-Authored-By: Claude <noreply@anthropic.com> * test(dogfood): e2e gate for the #3602 dimension-label read-scope leak Boots the real stack (HTTP → SecurityPlugin → AnalyticsServicePlugin auto-bridges → the plugin's real fetchRecordLabels closure) with a two-object owner-scoped fixture: a member owns a deal pointing at a vendor OWNED BY THE ADMIN. Grouping deals by vendor, resolving the grouped id to a NAME reads the vendor object — which the member cannot read. Asserts the member gets the vendor id RAW (the name does not leak) while the admin, who owns the vendor, still gets the resolved name — proving the fix scopes rather than blanks. Reverting the plugin's scope application reproduces the leak (member receives "Admin-Owned Vendor"). This is the only gate that exercises the plugin's real label closure; the service-analytics unit/integration tests fake fetchRecordLabels. Co-Authored-By: Claude <noreply@anthropic.com> --------- Co-authored-by: Jack Zhuang <277994282+os-zhuang@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com>
diff --git a/.changeset/analytics-label-read-scope.md b/.changeset/analytics-label-read-scope.md
@@ -0,0 +1,25 @@
+---
+"@objectstack/service-analytics": patch
+---
+
+fix(analytics): scope the dimension-label lookup to the referenced object's RLS (#3602)
+
+When a dataset groups by a `lookup`/`master_detail` dimension, analytics resolves
+the grouped FK ids to the related record's display name via a per-record read
+(`group by id`) dressed as an aggregate. That read carried **no read scope**, so
+it revealed related-record display names whenever the referenced object's RLS is
+stricter than the base object whose rows carry the id — a user could see a name
+the referenced object's own RLS would hide. (Same-object and looser-referenced
+cases were already safe because the ids come from the post-#3597 scoped
+aggregate; this closes the stricter-referenced case.)
+
+The label lookup now applies the **referenced object's own** read scope — bound
+to the request via the same `getReadScope` provider the aggregate path uses,
+composed with `$and` (never key-merge) so it can't be displaced by the id
+predicate. Fail-closed: if that object's scope can't be resolved, the dimension's
+labels are skipped (the raw id renders) rather than fetched unscoped. No behaviour
+change when no read-scope provider is configured.
+
+Internal `DimensionLabelDeps.fetchRecordLabels` gains an optional `scope` argument
+and `resolveDimensionLabels` an optional `resolveScope` resolver; both are
+service-analytics-internal (no spec/contract change).
diff --git a/packages/qa/dogfood/test/analytics-label-scope.dogfood.test.ts b/packages/qa/dogfood/test/analytics-label-scope.dogfood.test.ts
@@ -0,0 +1,98 @@
+// Copyright (c) 2026 ObjectStack. Licensed under the Apache-2.0 license.
+//
+// END-TO-END gate for #3602 residual 1 — "the dimension-label lookup is scoped
+// to the referenced object's RLS" — proven through the REAL stack: HTTP → REST
+// exec context → SecurityPlugin → AnalyticsServicePlugin auto-bridges → the
+// plugin's real fetchRecordLabels closure that ANDs the referenced object's
+// scope into the label read.
+//
+// The member owns a deal that points at a vendor OWNED BY THE ADMIN. Grouping
+// deals by vendor, the member's bucket carries the vendor id; resolving that id
+// to a NAME reads the vendor object, which the member cannot read. Before the
+// fix the member got the vendor's name (leak); after it, the vendor scope
+// excludes the admin's vendor so the raw id renders. The admin — who owns the
+// vendor — still sees the name, proving the fix SCOPES rather than blanks.
+//
+// This is the only gate that exercises the plugin's real label closure; the
+// service-analytics unit/integration tests fake fetchRecordLabels.
+
+import { describe, it, expect, beforeAll, afterAll } from 'vitest';
+import { bootStack, type VerifyStack } from '@objectstack/verify';
+import { labelScopeStack, labelScopeSecurity } from './fixtures/label-scope-fixture.js';
+
+const VENDOR_NAME = 'Admin-Owned Vendor';
+
+const dealsByVendor = {
+  name: 'deals_by_vendor',
+  label: 'Deals by vendor',
+  object: 'lbl_deal',
+  dimensions: [{ name: 'vendor', label: 'Vendor', field: 'vendor', type: 'lookup' }],
+  measures: [{ name: 'cnt', label: 'Count', aggregate: 'count' }],
+};
+
+describe('dogfood: dimension-label lookup is RLS-scoped to the referenced object (#3602)', () => {
+  let stack: VerifyStack;
+  let adminToken: string;
+  let memberToken: string;
+  let vendorId: string;
+
+  beforeAll(async () => {
+    stack = await bootStack(labelScopeStack as never, { security: labelScopeSecurity() });
+    adminToken = await stack.signIn();
+    memberToken = await stack.signUp('lbl-member@verify.test');
+
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    const ql = await stack.kernel.getServiceAsync<any>('objectql');
+    const sys = { context: { isSystem: true } };
+    const uid = async (email: string): Promise<string> => {
+      const u = await ql.findOne('sys_user', { where: { email }, ...sys });
+      if (!u?.id) throw new Error(`no sys_user for ${email}`);
+      return u.id as string;
+    };
+    const adminId = await uid('admin@objectos.ai');
+    const memberId = await uid('lbl-member@verify.test');
+
+    // Vendor owned by the admin — the member must NOT be able to read it.
+    const vendor = await ql.insert('lbl_vendor', { name: VENDOR_NAME, created_by: adminId }, sys);
+    vendorId = vendor.id as string;
+
+    // One deal per principal, BOTH pointing at the admin's vendor. Authored as
+    // system with an explicit owner so the owner policy binds deterministically
+    // (and the member's deal can reference a vendor it cannot itself read).
+    await ql.insert('lbl_deal', { name: 'member deal', amount: 10, vendor: vendorId, created_by: memberId }, sys);
+    await ql.insert('lbl_deal', { name: 'admin deal', amount: 20, vendor: vendorId, created_by: adminId }, sys);
+
+    // Preconditions: the member reads their deal but NOT the admin's vendor.
+    const deals = await stack.apiAs(memberToken, 'GET', '/data/lbl_deal');
+    expect(((await deals.json()).records ?? []).map((r: any) => r.name)).toEqual(['member deal']);
+    const vendors = await stack.apiAs(memberToken, 'GET', '/data/lbl_vendor');
+    expect((await vendors.json()).records ?? []).toHaveLength(0);
+  }, 90_000);
+
+  afterAll(async () => {
+    await stack?.stop();
+  });
+
+  async function vendorCellFor(token: string): Promise<unknown> {
+    const res = await stack.apiAs(token, 'POST', '/analytics/dataset/query', {
+      dataset: dealsByVendor,
+      selection: { dimensions: ['vendor'], measures: ['cnt'] },
+    });
+    expect(res.status).toBe(200);
+    const body = (await res.json()) as { rows?: Array<Record<string, unknown>> };
+    // Exactly one deal is visible to each principal → one vendor bucket.
+    expect(body.rows ?? []).toHaveLength(1);
+    return (body.rows ?? [])[0].vendor;
+  }
+
+  it('the member sees the vendor id RAW — the name it cannot read does not leak', async () => {
+    const cell = await vendorCellFor(memberToken);
+    expect(cell).toBe(vendorId);
+    expect(cell).not.toBe(VENDOR_NAME);
+  });
+
+  it('the admin — who owns the vendor — still sees the resolved name (scopes, not blanks)', async () => {
+    const cell = await vendorCellFor(adminToken);
+    expect(cell).toBe(VENDOR_NAME);
+  });
+});
diff --git a/packages/qa/dogfood/test/fixtures/label-scope-fixture.ts b/packages/qa/dogfood/test/fixtures/label-scope-fixture.ts
@@ -0,0 +1,86 @@
+// Copyright (c) 2026 ObjectStack. Licensed under the Apache-2.0 license.
+//
+// Two-object fixture for the #3602 dimension-label leak (residual 1).
+//
+// A dataset that groups by a LOOKUP dimension resolves each grouped FK id to the
+// related record's display NAME. That label read used to carry no read scope, so
+// it revealed the referenced record's name even when the reader could not read
+// that record — the leak fires whenever the base object's rows point at a
+// referenced record the reader's RLS hides.
+//
+// This fixture reproduces exactly that with zero dependence on org-scoping:
+//   • `lbl_vendor` — owner-scoped on `created_by`.
+//   • `lbl_deal`   — owner-scoped on `created_by`, with a `vendor` lookup.
+// A member owns a deal that points at a vendor OWNED BY THE ADMIN. The member can
+// read their deal (owner match) but NOT the vendor (owned by someone else), so a
+// deals-by-vendor dataset must show the vendor's raw id to the member and its
+// display name only to the admin who owns it.
+
+import { defineStack } from '@objectstack/spec';
+import { ObjectSchema, Field } from '@objectstack/spec/data';
+import { PermissionSetSchema, RLS, type PermissionSet } from '@objectstack/spec/security';
+import { SecurityPlugin, securityDefaultPermissionSets } from '@objectstack/plugin-security';
+
+export const Vendor = ObjectSchema.create({
+  name: 'lbl_vendor',
+  // [ADR-0090 D1] grandfather stamp — the gate under test is permission-set RLS,
+  // not owner-sharing.
+  sharingModel: 'public_read_write',
+  label: 'Vendor',
+  pluralLabel: 'Vendors',
+  fields: {
+    name: Field.text({ label: 'Name', required: true }),
+  },
+});
+
+export const Deal = ObjectSchema.create({
+  name: 'lbl_deal',
+  sharingModel: 'public_read_write',
+  label: 'Deal',
+  pluralLabel: 'Deals',
+  fields: {
+    name: Field.text({ label: 'Name', required: true }),
+    amount: Field.number({ label: 'Amount' }),
+    vendor: Field.lookup('lbl_vendor', { label: 'Vendor' }),
+  },
+});
+
+export const labelScopeStack = defineStack({
+  manifest: {
+    id: 'com.dogfood.label_scope',
+    namespace: 'lbl',
+    version: '0.0.0',
+    type: 'app',
+    name: 'Label Scope Fixture',
+    description: 'Deal → vendor lookup exercising the #3602 label read-scope leak.',
+  },
+  objects: [Vendor, Deal],
+});
+
+const MEMBER_SET = 'lbl_fixture_member';
+
+/**
+ * The fallback set a fresh member resolves to: CRUD on both objects (so requests
+ * reach the RLS layer rather than being denied by RBAC) plus an OWNER policy on
+ * each. The member therefore reads only rows they own — the precondition the
+ * label leak needs (their deal points at the admin's vendor).
+ */
+export const memberSet: PermissionSet = PermissionSetSchema.parse({
+  name: MEMBER_SET,
+  label: 'Label Fixture Member — owner-scoped',
+  objects: {
+    lbl_deal: { allowRead: true, allowCreate: true, allowEdit: true, allowDelete: true },
+    lbl_vendor: { allowRead: true, allowCreate: true, allowEdit: true, allowDelete: true },
+  },
+  rowLevelSecurity: [
+    RLS.ownerPolicy('lbl_deal', 'created_by'),
+    RLS.ownerPolicy('lbl_vendor', 'created_by'),
+  ],
+});
+
+export function labelScopeSecurity(): SecurityPlugin {
+  return new SecurityPlugin({
+    defaultPermissionSets: [...securityDefaultPermissionSets, memberSet],
+    fallbackPermissionSet: memberSet.name,
+  });
+}
diff --git a/packages/services/service-analytics/src/__tests__/dimension-labels.test.ts b/packages/services/service-analytics/src/__tests__/dimension-labels.test.ts
@@ -117,6 +117,71 @@ describe('resolveDimensionLabels', () => {
     expect(calls).toBe(1);
     expect(rows.map((r) => r.account)).toEqual(['Acme Corp', 'Acme Corp', 'Globex']);
   });
+
+  // ── #3602 — the label lookup must carry the REFERENCED object's read scope ──
+  describe('lookup label read scope (#3602)', () => {
+    it('passes the referenced object scope through to fetchRecordLabels', async () => {
+      let seenScope: unknown = 'UNSET';
+      let seenTarget: string | undefined;
+      const d = deps({
+        fetchRecordLabels: async (target, ids, scope) => {
+          seenTarget = target;
+          seenScope = scope;
+          return new Map<unknown, string>(ids.map((id) => [id, `name-${String(id)}`]));
+        },
+      });
+      const rows = [{ account: 'acc1', n: 1 }];
+      // resolveScope returns the referenced object's own RLS predicate.
+      await resolveDimensionLabels('task', [{ name: 'account', field: 'account' }], rows, d, (target) => {
+        expect(target).toBe('crm_account'); // the REFERENCED object, not the base
+        return { organization_id: 'org_A' };
+      });
+      expect(seenTarget).toBe('crm_account');
+      expect(seenScope).toEqual({ organization_id: 'org_A' });
+    });
+
+    it('fails CLOSED: when the scope cannot be resolved, the id is left raw (no unscoped fetch)', async () => {
+      let fetched = false;
+      const d = deps({
+        fetchRecordLabels: async (_t, ids) => {
+          fetched = true;
+          return new Map<unknown, string>(ids.map((id) => [id, 'LEAKED NAME']));
+        },
+      });
+      const rows = [{ account: 'acc1', n: 1 }];
+      await resolveDimensionLabels('task', [{ name: 'account', field: 'account' }], rows, d, () => {
+        throw new Error('security service unavailable');
+      });
+      // The label fetch must NOT have run, and the raw id must survive.
+      expect(fetched).toBe(false);
+      expect(rows).toEqual([{ account: 'acc1', n: 1 }]);
+    });
+
+    it('a select dimension is unaffected by scope resolution (no referenced object)', async () => {
+      let scopeCalls = 0;
+      const rows = [{ status: 'backlog', n: 1 }];
+      await resolveDimensionLabels('task', [{ name: 'status', field: 'status' }], rows, deps(), () => {
+        scopeCalls++;
+        return undefined;
+      });
+      expect(scopeCalls).toBe(0); // scope is only resolved for lookup/master_detail dims
+      expect(rows).toEqual([{ status: 'Backlog', n: 1 }]);
+    });
+
+    it('no resolver (no security configured) → unscoped fetch, unchanged behaviour', async () => {
+      let seenScope: unknown = 'UNSET';
+      const d = deps({
+        fetchRecordLabels: async (_t, ids, scope) => {
+          seenScope = scope;
+          return new Map<unknown, string>(ids.map((id) => [id, 'Acme Corp']));
+        },
+      });
+      const rows = [{ account: 'acc1', n: 1 }];
+      await resolveDimensionLabels('task', [{ name: 'account', field: 'account' }], rows, d /* no resolveScope */);
+      expect(seenScope).toBeUndefined();
+      expect(rows).toEqual([{ account: 'Acme Corp', n: 1 }]);
+    });
+  });
 });
 
 describe('formatDateBucket', () => {
diff --git a/packages/services/service-analytics/src/__tests__/query-dataset.test.ts b/packages/services/service-analytics/src/__tests__/query-dataset.test.ts
@@ -393,4 +393,52 @@ describe('AnalyticsService.queryDataset', () => {
     expect(result.totals[1].dimensions).toEqual([]);
     expect(result.drillRawTotals[1]).toEqual([{}]);
   });
+
+  // ── #3602 — the lookup label read is scoped to the REFERENCED object's RLS ──
+  it('threads the referenced object read scope (not the base object) into the label fetch', async () => {
+    const byAccount = DatasetSchema.parse({
+      name: 'sales_acct_scoped', label: 'Sales', object: 'opportunity', include: [],
+      dimensions: [{ name: 'account', field: 'account', type: 'lookup', label: 'Account' }],
+      measures: [{ name: 'revenue', aggregate: 'sum', field: 'amount' }],
+    });
+    const labelScopes: Array<{ target: string; scope: unknown }> = [];
+    const svc = new AnalyticsService({
+      queryCapabilities: () => ({ nativeSql: true, objectqlAggregate: false, inMemory: false }),
+      executeRawSql: async () => [{ account: 'acc1', revenue: 1000 }],
+      // Per-object scope: opportunity and crm_account get DIFFERENT predicates, so
+      // the assertion proves the label fetch used the referenced object's scope,
+      // not the base object's.
+      getReadScope: (object, ctx?: ExecutionContext) => {
+        if (!ctx?.tenantId) return undefined;
+        return object === 'crm_account'
+          ? { organization_id: ctx.tenantId, is_public: true }
+          : { organization_id: ctx.tenantId };
+      },
+      labelResolver: {
+        getObjectFields: (obj) => ({
+          opportunity: { account: { type: 'lookup', reference: 'crm_account' } },
+          crm_account: { name: { type: 'text' } },
+        } as Record<string, Record<string, { type?: string; reference?: string }>>)[obj],
+        fetchRecordLabels: async (target, ids, scope) => {
+          labelScopes.push({ target, scope });
+          const m = new Map<unknown, string>();
+          for (const id of ids) m.set(id, `name-${String(id)}`);
+          return m;
+        },
+      },
+    });
+
+    const result = await svc.queryDataset(
+      byAccount,
+      { dimensions: ['account'], measures: ['revenue'] },
+      { tenantId: 'org_A' } as ExecutionContext,
+    ) as any;
+
+    // The label fetch ran against the referenced object, carrying ITS scope.
+    expect(labelScopes).toEqual([
+      { target: 'crm_account', scope: { organization_id: 'org_A', is_public: true } },
+    ]);
+    // And the label actually resolved (end-to-end sanity).
+    expect(result.rows).toEqual([{ account: 'name-acc1', revenue: 1000 }]);
+  });
 });
diff --git a/packages/services/service-analytics/src/analytics-service.ts b/packages/services/service-analytics/src/analytics-service.ts
@@ -604,14 +604,23 @@ export class AnalyticsService implements IAnalyticsService {
         .filter((d) => !!d.field)
         .map((d) => ({ name: d.name, field: d.field, type: d.type, dateGranularity: d.dateGranularity }));
       if (dims.length) {
+        // #3602 — bind the referenced object's read scope to THIS request so the
+        // label lookup (a per-record read of the related object) cannot surface a
+        // record the referenced object's RLS would hide. Same provider the
+        // aggregate path uses; `undefined` when no provider is configured, in
+        // which case labels fetch unscoped exactly as before.
+        const provider = this.readScopeProvider;
+        const resolveScope = provider
+          ? (targetObject: string) => provider(targetObject, context)
+          : undefined;
         try {
-          await resolveDimensionLabels(dataset.object, dims, result.rows, this.labelResolver);
+          await resolveDimensionLabels(dataset.object, dims, result.rows, this.labelResolver, resolveScope);
           // Totals rows (#1753) carry dimension values too (a row subtotal is
           // keyed by its row bucket) — resolve each grouping's own subset.
           for (const total of result.totals ?? []) {
             const subset = dims.filter((d) => total.dimensions.includes(d.name));
             if (subset.length) {
-              await resolveDimensionLabels(dataset.object, subset, total.rows, this.labelResolver);
+              await resolveDimensionLabels(dataset.object, subset, total.rows, this.labelResolver, resolveScope);
             }
           }
         } catch (e) {
diff --git a/packages/services/service-analytics/src/dimension-labels.ts b/packages/services/service-analytics/src/dimension-labels.ts
diff --git a/packages/services/service-analytics/src/plugin.ts b/packages/services/service-analytics/src/plugin.ts
