gerbeur/api/routes/files.ts

/**
 * Serves file dumps' uploads to the players, viewers and download links of the
 * frontend, which builds the URL with `dumpFileUrl`.
 *
 * - `GET /api/files/:dumpId?v=&token=` — public; a private dump's file needs
 *   its poster's media `token`. Serves the file, inline when that is safe and
 *   as a download otherwise, with byte ranges.
 *
 * @module
 */

import { Router } from "@oak/oak";
import { APIException } from "../lib/api-exception.ts";
import { DUMPS_DIR } from "../config.ts";
import { authenticateMedia } from "../lib/auth.ts";
import { serveFile } from "../lib/serve-file.ts";
import { dumpFileVersion, getDump } from "../services/dump-service.ts";

/**
 * Mounted by `api/main.ts` behind the error, Open Graph and CORS middleware.
 * An `APIException` thrown by a handler becomes the JSON error response there.
 */
const router: Router = new Router({ prefix: "/api/files" });

/**
 * Builds a `Content-Disposition` value for a file called `name`. `Headers.set`
 * throws on any character above U+00FF, so `filename` gets an ASCII copy of the
 * name and `filename*` (RFC 5987) the full one, percent-encoded.
 */
function contentDisposition(
  type: "inline" | "attachment",
  name: string,
): string {
  const asciiFallback = name
    .replace(/["\\]/g, "_")
    .replace(/[^\x20-\x7e]/g, "_");
  return `${type}; filename="${asciiFallback}"; filename*=UTF-8''${
    encodeURIComponent(name)
  }`;
}

/**
 * Matches the types served inline. An upload's type is whatever the uploader's
 * browser claimed, and the file comes from the app's own origin, where the
 * session token lives. So only types a browser shows without running code in
 * them are inline, and anything else (HTML, SVG, XML…) downloads. `<img>`,
 * `<audio>` and `fetch()` ignore `Content-Disposition`, so the players and
 * previews still work.
 */
const INLINE_MIME_RE =
  /^(image\/(png|jpeg|gif|webp|avif|bmp)|audio\/[\w.+-]+|video\/[\w.+-]+|application\/pdf|text\/plain)$/;

function isInlineSafe(mime: string): boolean {
  return INLINE_MIME_RE.test(mime.split(";")[0].trim().toLowerCase());
}

router.get("/:dumpId", async (ctx) => {
  const { dumpId } = ctx.params;

  // The id names a file on disk: nothing but an id's characters gets through.
  if (!/^[0-9a-f-]{36}$/.test(dumpId)) {
    throw new APIException(400, "Invalid dump ID");
  }

  // Media elements can't send an Authorization header, so the viewer's media
  // token comes in the query string.
  const token = ctx.request.url.searchParams.get("token");
  const viewerId = token ? await authenticateMedia(token) : null;
  const dump = getDump(dumpId, viewerId ?? undefined);

  if (dump.kind !== "file" || !dump.fileMime || !dump.fileName) {
    throw new APIException(404, "No file for this dump");
  }

  const path = `${DUMPS_DIR}/${dumpId}`;
  const inline = isInlineSafe(dump.fileMime);
  ctx.response.headers.set(
    "Content-Disposition",
    contentDisposition(inline ? "inline" : "attachment", dump.fileName),
  );
  ctx.response.headers.set("X-Content-Type-Options", "nosniff");
  // Belt and braces for a file opened directly: no scripts, no plugins, no
  // same-origin access. Chrome won't render a PDF in a sandboxed document, and
  // its viewer never runs a PDF's scripts anyway, so PDFs go without.
  if (dump.fileMime !== "application/pdf") {
    ctx.response.headers.set(
      "Content-Security-Policy",
      "sandbox; default-src 'none'; img-src 'self'; media-src 'self'; style-src 'unsafe-inline'",
    );
  }
  // The URL may carry a media token; don't pass it on to wherever the file
  // links.
  ctx.response.headers.set("Referrer-Policy", "no-referrer");

  // A URL naming the file's current version (dumpFileVersion) always serves the
  // same bytes, so it is cached for good; any other is revalidated against the
  // ETag. A private dump's bytes must not sit in a shared cache.
  const scope = dump.isPrivate ? "private" : "public";
  const current = ctx.request.url.searchParams.get("v") ===
    dumpFileVersion(dump);
  const served = await serveFile(ctx, path, {
    mime: dump.fileMime,
    cacheControl: current
      ? `${scope}, max-age=31536000, immutable`
      : `${scope}, no-cache`,
    ranges: true,
  });
  if (!served) {
    throw new APIException(404, "File not found");
  }
});

export default router;