Files
Ekklesia/apps/api/src/module-http-routes.ts
T
ekklesiaandClaude Opus 5 e36485bf99 feat: let a module route stream an image-sized upload, explicitly and bounded
ADR 0017 capped a module route at 1 MiB and said large media belonged to a
future storage design. That design arrived — ADR 0030 gave modules a bucket,
ADR 0031 took Ekklesia out of the read path — and it did not solve the write
path: signed write URLs were deliberately declined, so the only way bytes
reach a bucket is a module's own route, and a photograph is not 1 MiB.

Extend the route contract rather than adding a second HTTP system. A route
declares `bodyMode`: `buffered` (the default, and what every existing route
already meant) keeps the 1 MiB ceiling and materializes the body before the
handler runs; `streamed` raises the ceiling to 32 MiB and counts bytes as the
handler consumes them, holding nothing. The mode is explicit rather than
inferred from the size, because inferring it would mean an author who raised a
limit silently changed how their own handler receives its body and when a 413
can reach their own code.

32 MiB holds one 25 MiB image plus bounded overhead and is also the platform's
absolute ceiling: no route in any mode may declare more, and no unbounded
option exists in the contract. The unit is one file per request — twenty
images are twenty individually bounded requests, and nothing here can express
a 500 MiB aggregate whose failure would waste everything already sent.

Replace the buffering dependency with the platform's own limiter. A declared
Content-Length over the limit is refused synchronously, before authorization
and before a byte is read. Buffered mode now counts what actually arrives
rather than trusting the header, which closes a gap the previous
implementation had. Streamed mode wraps the body in a counting transform;
on overflow the stream errors, the source is cancelled, and the response is
replaced with the same sanitized 413 even if the handler swallowed the error —
a limit a handler can swallow is not a limit. An unread body is cancelled so
an abandoned upload releases its socket, and an abort stays an ordinary failed
request rather than becoming a 413.

Ordering is unchanged and now strictly better: in streamed mode no byte is
read until the handler reads it, and the handler runs only after authorization
succeeds, so an unauthorized caller cannot push an upload through at all.

The live suite uploads a real 25 MiB body over a real socket end to end.

Co-Authored-By: Claude Opus 5 <[email protected]>
2026-08-27 15:57:44 -05:00

285 lines
14 KiB
TypeScript

import {
EMPTY_MODULE_CONFIGURATION,
EMPTY_MODULE_STORAGE,
type ModuleId,
type ModuleRouteHandlerContext,
type ModuleRuntimeServices,
type RegisteredModuleHttpRoute,
} from "@ekklesia/contracts";
import type {
ModuleConfigurationLookup,
ModuleRealtimeRegistry,
ModuleStorageLookup,
} from "@ekklesia/core/modules";
import type { Context, Hono } from "hono";
import {
type AuthorizationMiddlewareOptions,
getAuthorizedPrincipal,
requirePermission,
requireTrustedOrigin,
} from "./authorization-middleware.js";
import { platformBodyLimit } from "./body-limits.js";
const MODULE_ROUTE_PREFIX = "/api/v1/modules";
const STATE_CHANGING_METHODS: ReadonlySet<string> = new Set(["DELETE", "PATCH", "POST", "PUT"]);
const STRIPPED_CORS_RESPONSE_HEADERS = [
"access-control-allow-credentials",
"access-control-allow-headers",
"access-control-allow-methods",
"access-control-allow-origin",
"access-control-expose-headers",
"access-control-max-age",
];
export interface ModuleHttpRoutesOptions extends AuthorizationMiddlewareOptions {
readonly trustedOrigins: ReadonlySet<string>;
/** The same frozen services object the started module runtime's `start`/`stop` hooks already
* received for this API runtime — never a fresh one built per request, and never touched for a
* denied, trusted-origin-rejected, or `OPTIONS` request, since a handler is only ever invoked
* after authorization succeeds. */
readonly services: ModuleRuntimeServices;
/**
* Per-module frozen configuration snapshots — the exact map the module runtime started from
* (ADR 0029), so a handler and that module's `start`/`stop` hooks receive the identical object
* rather than two independently resolved values. A module with no entry receives
* `EMPTY_MODULE_CONFIGURATION`, so `context.config` is always present.
*
* Unlike `services`, which is one shared object, this is keyed by module id: a handler can only
* ever reach its own module's configuration, because that is the only entry looked up for it.
*/
readonly configurations: ModuleConfigurationLookup;
/**
* Per-module scoped object storage — the exact map the module runtime started from (ADR 0030), so
* a handler and that module's lifecycle hooks share one facade rather than two independently
* resolved ones.
*
* Keyed by module id, like `configurations` and unlike `services`: a handler can only ever reach
* its own module's storage, because that is the only entry looked up for it.
*/
readonly storages: ModuleStorageLookup;
/** The same started registry whose scoped publishers lifecycle hooks receive. */
readonly realtime: ModuleRealtimeRegistry;
}
/** Never carries the offending permission id, route path, module id, or any other configuration
* detail — a forged or inconsistent route snapshot is a startup-time composition bug, not
* something to describe in detail to whatever eventually logs it. */
export class ModuleRouteConfigurationError extends Error {
constructor() {
super("A published module HTTP route references a permission that is not registered.");
this.name = "ModuleRouteConfigurationError";
this.stack = `${this.name}: ${this.message}`;
Object.freeze(this);
}
}
/**
* The root a module's own HTTP API is mounted at — the one place that URL shape is constructed.
*
* Exported because a browser host has to resolve `context.api.request("records")`
* (`ModuleUiApiClient`, `@ekklesia/module-ui-sdk`) to a real Ekklesia URL, and a module UI must
* never build that prefix itself: it is Ekklesia's private deployment detail, and a module that
* hardcoded it could not be hosted anywhere else (ADR 0022). The server hands the resolved base to
* the host through the module UI inventory (`./module-ui-inventory.js`) rather than letting the
* browser re-spell this template, so this prefix can change without touching either.
*
* Percent-encoded for the same reason `moduleUiArtifactUrl` is: a value that somehow did not come
* from a validated manifest can only ever become one segment's worth of text, never a new segment.
* A real module id (lowercase kebab-case, `moduleIdSchema`) is unaffected by the encoding.
*/
export function moduleApiBaseUrl(moduleId: string): string {
return `${MODULE_ROUTE_PREFIX}/${encodeURIComponent(moduleId)}`;
}
/** "/" is the module's own root and maps to the bare module prefix, never a required trailing
* slash; every other local path appends directly. Root and non-root therefore never collide, and
* `/api/v1/modules/<module-id>/` (the root path with a trailing slash) is never produced here. */
function moduleRoutePath(moduleId: ModuleId, routePath: string): string {
const base = moduleApiBaseUrl(moduleId);
return routePath === "/" ? base : `${base}${routePath}`;
}
function sanitizedModuleHandlerFailure(c: Context): Response {
c.header("Cache-Control", "no-store");
return c.json({ code: "MODULE_HANDLER_FAILED" }, 500);
}
/**
* The one shared `Vary` parsing/canonicalization primitive: splits and trims comma-separated
* tokens and de-duplicates them case-insensitively — `Vary` lists header *names*, which are
* themselves case-insensitive, so a module's own `origin` must collapse with a later `Origin`
* rather than appear as a separate, redundant entry — while preserving first-seen casing and
* relative order. Never returns early on `*`; callers decide what a `*` token means for their own
* policy (see `mergedVaryHeader` and `canonicalizeVaryHeader` below, the two callers that build on
* this instead of re-implementing parsing/dedup independently).
*/
function parseVaryTokens(existingVary: string | null): Map<string, string> {
const canonicalByLowercase = new Map<string, string>();
for (const token of (existingVary ?? "").split(",")) {
const trimmed = token.trim();
if (trimmed.length === 0) {
continue;
}
const key = trimmed.toLowerCase();
if (!canonicalByLowercase.has(key)) {
canonicalByLowercase.set(key, trimmed);
}
}
return canonicalByLowercase;
}
/** A literal `*` (RFC 9110 §12.5.5) must never be combined with any other value; since `*` already
* means "may vary on anything," a strictly broader signal than any explicit list, its presence
* collapses the result to `*` alone rather than producing an invalid combination. */
function serializeVaryTokens(tokens: ReadonlyMap<string, string>): string {
if (tokens.has("*")) {
return "*";
}
return [...tokens.values()].join(", ");
}
/**
* Used only for a module's own response: always emits the canonical `Origin`/`Cookie` casing
* regardless of how (or whether) a module already listed them, since every module response is
* forced to vary on both (see `sanitizeModuleResponse`) — this is a deliberate product policy for
* module responses specifically, not a generic property every response through the platform's CORS
* boundary should have forced onto it (an `OPTIONS` preflight, for one, has no `Cookie` concept at
* all).
*/
export function mergedVaryHeader(existingVary: string | null): string {
const tokens = parseVaryTokens(existingVary);
if (tokens.has("*")) {
return "*";
}
tokens.set("origin", "Origin");
tokens.set("cookie", "Cookie");
return serializeVaryTokens(tokens);
}
/**
* Used at the app-level CORS boundary (`apps/api/src/app.ts`), which wraps every response —
* `OPTIONS` preflights and Core's own routes included, not only module ones — after Hono's own
* `cors()` has already appended its own `Vary: Origin`. Unlike `mergedVaryHeader`, this never
* injects a token that was not already present; it only de-duplicates and collapses `*`, since a
* generic boundary must not assume every response wants `Origin`/`Cookie` forced onto it.
*/
export function canonicalizeVaryHeader(existingVary: string): string {
return serializeVaryTokens(parseVaryTokens(existingVary));
}
/**
* The platform owns CORS and caching for every module route: strips any CORS-control header a
* handler set (defense in depth — the outer `cors()` middleware already computes the authoritative
* values for the actual response) and forces `Cache-Control: no-store` plus a `Vary` that always
* includes `Origin` and `Cookie`, since even a public-baseline handler receives an authorized
* principal and may personalize its response. A separate cache-policy contract is future work;
* this is the conservative default until then. A module's own legitimate `Vary` values (if any)
* are preserved alongside the forced ones. Never reads or buffers `response.body` — it is passed
* through to the new `Response` unread, so a streaming body is neither consumed nor blocked here.
*/
function sanitizeModuleResponse(response: Response): Response {
const headers = new Headers(response.headers);
for (const headerName of STRIPPED_CORS_RESPONSE_HEADERS) {
headers.delete(headerName);
}
headers.set("Cache-Control", "no-store");
headers.set("Vary", mergedVaryHeader(headers.get("Vary")));
return new Response(response.body, {
headers,
status: response.status,
statusText: response.statusText,
});
}
/**
* Mounts `StartedModuleRuntime`'s published, immutable route snapshot into `app`, in publication
* order — never sorted or mutated, and never invoking a handler while mounting. Every route is
* protected by the exact same `requirePermission` middleware Core's own routes use, with
* `route.permission` as the only permission id ever consulted and its baseline coming exclusively
* from `options.registry`; a state-changing method (`POST`/`PUT`/`PATCH`/`DELETE`) additionally
* requires `requireTrustedOrigin` before authorization ever runs. `GET` inherits Hono's own
* implicit `HEAD` handling, which re-enters this exact same middleware chain and strips the
* response body afterward — there is no separate `HEAD` registration or permission path.
* `route.handler` is invoked only after authorization succeeds, with the raw Fetch `Request` and a
* frozen context carrying exactly `config`, `moduleId`, `params`, `principal`, `realtime`,
* `services`, and `storage` — never a Hono `Context`, authorization store, raw provider, or
* deployment-wide configuration. `services.database` is the module's only reachable database
* surface, and the whole context is unreachable to anything short of a successfully authorized
* handler invocation (a denied request, a rejected trusted-origin check, and `OPTIONS` never reach
* this handler at all).
*
* Because the routes and registry both derive from the same discovered inventory
* (`prepareStartupModuleInventory`), every registered route permission should already be
* registered; this loop still checks that invariant explicitly and throws
* `ModuleRouteConfigurationError` before mounting anything further if it ever does not hold, so a
* forged or hand-built route snapshot fails app construction instead of silently mounting an
* unenforceable route. `requirePermission`'s own per-request registry lookup remains the live,
* defense-in-depth enforcement regardless.
*/
export function mountModuleHttpRoutes(
app: Hono,
routes: readonly RegisteredModuleHttpRoute[],
options: ModuleHttpRoutesOptions,
): void {
const middlewareOptions: AuthorizationMiddlewareOptions = {
evaluator: options.evaluator,
registry: options.registry,
runtime: options.runtime,
};
const requireOrigin = requireTrustedOrigin(options.trustedOrigins);
for (const route of routes) {
if (!options.registry.has(route.permission)) {
throw new ModuleRouteConfigurationError();
}
const requireRoutePermission = requirePermission(middlewareOptions, route.permission);
// The route's own conscious body contract (`route.maxBodyBytes` — 0 means no body accepted at
// all) is enforced before permission evaluation ever runs, so a request *declared* oversized
// never reaches authorization work, let alone the handler or the module database. In
// `streamed` mode no body byte is read until the handler reads it, and the handler runs only
// after authorization succeeds, so an unauthorized caller can never push an upload through
// (ADR 0042).
const routeBodyLimit = platformBodyLimit(route.maxBodyBytes, route.bodyMode);
const path = moduleRoutePath(route.moduleId, route.path);
const finalHandler = async (c: Context): Promise<Response> => {
const context: ModuleRouteHandlerContext = Object.freeze({
config: options.configurations.get(route.moduleId) ?? EMPTY_MODULE_CONFIGURATION,
moduleId: route.moduleId,
storage: options.storages.get(route.moduleId) ?? EMPTY_MODULE_STORAGE,
params: Object.freeze({ ...c.req.param() }),
principal: getAuthorizedPrincipal(c),
realtime: options.realtime.publisherFor(route.moduleId),
services: options.services,
});
let response: Response;
try {
response = await route.handler(c.req.raw, context);
} catch {
return sanitizedModuleHandlerFailure(c);
}
if (!(response instanceof Response)) {
return sanitizedModuleHandlerFailure(c);
}
return sanitizeModuleResponse(response);
};
// Hono's `on()` overloads resolve by literal call arity, not by the runtime length of a spread
// array — each branch below is therefore a distinct, fixed-arity call rather than one call
// spreading a variable-length handlers array.
if (STATE_CHANGING_METHODS.has(route.method)) {
app.on(
route.method,
path,
requireOrigin,
routeBodyLimit,
requireRoutePermission,
finalHandler,
);
} else {
app.on(route.method, path, routeBodyLimit, requireRoutePermission, finalHandler);
}
}
}