fix: keep a document's frames, scripts and styles from reaching MemoryStick's own page

.github/scripts/check-document-filter.py

@@ -0,0 +1,34 @@
+#!/usr/bin/env python3
+"""Check that render() in dist/app.js still drops every element that could act
+on MemoryStick's own page.
+
+comrak keeps a document's raw HTML as written, and the inert <template> pass
+in render() is the only place these elements are removed. This fails if one
+of them is missing from DROPPED_ELEMENTS, or if render() stops using it.
+"""
+
+import re
+import sys
+from pathlib import Path
+
+REQUIRED = {"meta", "link", "iframe", "frame", "frameset", "object", "embed", "script", "base", "style"}
+# Ordinary raw HTML a document is allowed to keep, as on GitHub.
+KEPT = {"kbd", "details", "summary", "sub", "sup", "svg", "img", "a"}
+
+app = (Path(__file__).resolve().parents[2] / "dist" / "app.js").read_text(encoding="utf-8")
+
+errors = []
+match = re.search(r"^const DROPPED_ELEMENTS = '([^']*)';$", app, flags=re.M)
+if not match:
+    errors.append("dist/app.js: no `const DROPPED_ELEMENTS = '...';` line")
+else:
+    dropped = {name.strip() for name in match.group(1).split(",")}
+    if missing := REQUIRED - dropped:
+        errors.append(f"DROPPED_ELEMENTS no longer drops: {', '.join(sorted(missing))}")
+    if wrongly := KEPT & dropped:
+        errors.append(f"DROPPED_ELEMENTS drops ordinary HTML: {', '.join(sorted(wrongly))}")
+if ".querySelectorAll(DROPPED_ELEMENTS).forEach((el) => el.remove());" not in app:
+    errors.append("dist/app.js: render() no longer removes DROPPED_ELEMENTS from the document")
+
+print("\n".join(errors) or f"OK: render() drops all {len(REQUIRED)} elements")
+sys.exit(1 if errors else 0)

.github/workflows/ci.yml

@@ -142,3 +142,21 @@ jobs:
 
       - name: Every theme defines the colours dist/ uses
         run: python3 .github/scripts/check-colony-themes.py
+
+  # render() in dist/app.js is the only place a document's scripts, frames
+  # and style sheets are removed (comrak keeps raw HTML as written).
+  document-filter:
+    name: Document filter
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        shell: bash
+        working-directory: .
+
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          persist-credentials: false
+
+      - name: render() drops every element that could act on the page
+        run: python3 .github/scripts/check-document-filter.py

README.md

@@ -147,8 +147,9 @@ any other file, or a network share by its address, such as
 `\\server\share` on Windows. It does not upload, copy or keep these
 files.
 
-Documents can contain HTML, which is displayed, but no script contained in a
-document is ever run. The window always shows MemoryStick's own page: nothing
+Documents can contain HTML, which is displayed, apart from scripts, frames,
+embedded objects and style sheets, which are left out: no script contained in
+a document is ever run. The window always shows MemoryStick's own page: nothing
 in a document can navigate it to another page or file. Web and email links
 open in your default browser or mail app, and only when you click them;
 links to a place in the document scroll to it, and other links do nothing.

dist/app.js

@@ -304,19 +304,27 @@ document.addEventListener('click', (e) => {
 }, true);
 
 // ===== Rendering =====
+// The elements render() drops from a document. .github/scripts/
+// check-document-filter.py fails CI if one of them goes missing.
+const DROPPED_ELEMENTS = 'meta, link, iframe, frame, frameset, object, embed, script, base, style';
+
 // The page never names a file: main.rs reads only the document the user
 // opened, through the Open dialog, a drop or the command line.
 function render(result) {
   currentFilePath = result.file_path;
   currentDir = dirname(currentFilePath);
 
-  // Parsed in an inert <template> first, so that the document's <meta>
-  // elements (a refresh tag navigates the window) and <link> elements
-  // (preconnect and prefetch hints reach the network) are dropped before
-  // the page ever sees them.
+  // Parsed in an inert <template> first, so that the elements a document
+  // could use to act on MemoryStick's own page are dropped before the page
+  // ever sees them: <meta> (a refresh tag navigates the window), <link>
+  // (preconnect and prefetch hints reach the network), frames, <object> and
+  // <embed> (an <iframe srcdoc> shares the page's origin and could load its
+  // scripts), <script>, <base> (it would move every relative link) and
+  // <style> (its rules would restyle the interface, not only the document).
+  // comrak keeps raw HTML as it is, so this is the one place they are removed.
   const doc = document.createElement('template');
   doc.innerHTML = result.html;
-  doc.content.querySelectorAll('meta, link').forEach((el) => el.remove());
+  doc.content.querySelectorAll(DROPPED_ELEMENTS).forEach((el) => el.remove());
   $content.replaceChildren(doc.content);
   $main.classList.add('has-content');
   // A document takes the window back from Preferences.

dist/styles.css

@@ -202,10 +202,14 @@ body.drag-over #drop-hint .hint-title { color: var(--colony-accent-blue); }
 
 /* ===== Markdown content =====
    .markdown-body is the document, and the preview in Preferences. */
+/* contain: paint keeps a document inside its own box: an element with an
+   inline position: fixed is placed and clipped within #content, so it
+   cannot be drawn over the toolbar or the rest of the interface. */
 #content {
   max-width: 920px;
   margin: 0 auto;
   padding: 40px 60px 100px;
+  contain: paint;
 }
 .markdown-body {
   line-height: 1.65;

src-tauri/src/markdown.rs

@@ -15,6 +15,9 @@ pub fn render(markdown: &str) -> String {
     ext.header_id_prefix = Some("h-".to_string());
     ext.front_matter_delimiter = Some("---".to_string());
     opts.parse.smart = true;
+    // Raw HTML is kept as written (kbd, details, sub...). render() in app.js
+    // drops the elements that could act on the app's own page, such as
+    // scripts, frames and style sheets, before the document is shown.
     opts.render.r#unsafe = true;
 
     markdown_to_html(markdown, &opts)