vscode/src/vs/base/browser/markdownRenderer.ts
Matt Bierner 474d4951d8
Switch to dompurify for sanitizing markdown content (#131950)
* Switch to dompurify for sanitizing markdown content

Switches us from using `insane` to instead use `dompurify`, which seems to be better maintained and also has some nice features, such as built-in trusted types support

I've tried to port over our existing sanitizer settings as best as possible, but there's not always a 1:1 mapping between how insane works and how dompurify does. I'd like to get this change in early in the iteration to catch potential regressions

* Remove logging and renaming param

* Move dompurify to browser layer

* Fixing tests and how we check valid attributes

* Allow innerhtml in specific files

* Use isEqualNode instead of checking innerHTML directly

innerHTML can return different results on different browsers. Use `isEqualNode` instead

* Reapply fix for trusted types

* Enable ALLOW_UNKNOWN_PROTOCOLS

I beleive this is required since we allow links to commands and loading images over remote

* in -> of

* Fix check of protocol

* Enable two more safe tags
2021-09-03 12:17:02 -07:00

426 lines
14 KiB
TypeScript

/*---------------------------------------------------------------------------------------------
* Copyright (c) Microsoft Corporation. All rights reserved.
* Licensed under the MIT License. See License.txt in the project root for license information.
*--------------------------------------------------------------------------------------------*/
import * as DOM from 'vs/base/browser/dom';
import * as dompurify from 'vs/base/browser/dompurify/dompurify';
import { DomEmitter } from 'vs/base/browser/event';
import { createElement, FormattedTextRenderOptions } from 'vs/base/browser/formattedTextRenderer';
import { StandardMouseEvent } from 'vs/base/browser/mouseEvent';
import { renderLabelWithIcons } from 'vs/base/browser/ui/iconLabel/iconLabels';
import { onUnexpectedError } from 'vs/base/common/errors';
import { Event } from 'vs/base/common/event';
import { IMarkdownString, parseHrefAndDimensions, removeMarkdownEscapes } from 'vs/base/common/htmlContent';
import { markdownEscapeEscapedIcons } from 'vs/base/common/iconLabels';
import { defaultGenerator } from 'vs/base/common/idGenerator';
import * as marked from 'vs/base/common/marked/marked';
import { parse } from 'vs/base/common/marshalling';
import { FileAccess, Schemas } from 'vs/base/common/network';
import { cloneAndChange } from 'vs/base/common/objects';
import { resolvePath } from 'vs/base/common/resources';
import { escape } from 'vs/base/common/strings';
import { URI } from 'vs/base/common/uri';
export interface MarkedOptions extends marked.MarkedOptions {
baseUrl?: never;
}
export interface MarkdownRenderOptions extends FormattedTextRenderOptions {
codeBlockRenderer?: (modeId: string, value: string) => Promise<HTMLElement>;
asyncRenderCallback?: () => void;
baseUrl?: URI;
}
/**
* Low-level way create a html element from a markdown string.
*
* **Note** that for most cases you should be using [`MarkdownRenderer`](./src/vs/editor/browser/core/markdownRenderer.ts)
* which comes with support for pretty code block rendering and which uses the default way of handling links.
*/
export function renderMarkdown(markdown: IMarkdownString, options: MarkdownRenderOptions = {}, markedOptions: MarkedOptions = {}): HTMLElement {
const element = createElement(options);
const _uriMassage = function (part: string): string {
let data: any;
try {
data = parse(decodeURIComponent(part));
} catch (e) {
// ignore
}
if (!data) {
return part;
}
data = cloneAndChange(data, value => {
if (markdown.uris && markdown.uris[value]) {
return URI.revive(markdown.uris[value]);
} else {
return undefined;
}
});
return encodeURIComponent(JSON.stringify(data));
};
const _href = function (href: string, isDomUri: boolean): string {
const data = markdown.uris && markdown.uris[href];
if (!data) {
return href; // no uri exists
}
let uri = URI.revive(data);
if (isDomUri) {
if (href.startsWith(Schemas.data + ':')) {
return href;
}
// this URI will end up as "src"-attribute of a dom node
// and because of that special rewriting needs to be done
// so that the URI uses a protocol that's understood by
// browsers (like http or https)
return FileAccess.asBrowserUri(uri).toString(true);
}
if (URI.parse(href).toString() === uri.toString()) {
return href; // no transformation performed
}
if (uri.query) {
uri = uri.with({ query: _uriMassage(uri.query) });
}
return uri.toString();
};
// signal to code-block render that the
// element has been created
let signalInnerHTML: () => void;
const withInnerHTML = new Promise<void>(c => signalInnerHTML = c);
const renderer = new marked.Renderer();
renderer.image = (href: string, title: string, text: string) => {
let dimensions: string[] = [];
let attributes: string[] = [];
if (href) {
({ href, dimensions } = parseHrefAndDimensions(href));
href = _href(href, true);
try {
const hrefAsUri = URI.parse(href);
if (options.baseUrl && hrefAsUri.scheme === Schemas.file) { // absolute or relative local path, or file: uri
href = resolvePath(options.baseUrl, href).toString();
}
} catch (err) { }
attributes.push(`src="${href}"`);
}
if (text) {
attributes.push(`alt="${text}"`);
}
if (title) {
attributes.push(`title="${title}"`);
}
if (dimensions.length) {
attributes = attributes.concat(dimensions);
}
return '<img ' + attributes.join(' ') + '>';
};
renderer.link = (href, title, text): string => {
// Remove markdown escapes. Workaround for https://github.com/chjj/marked/issues/829
if (href === text) { // raw link case
text = removeMarkdownEscapes(text);
}
href = _href(href, false);
if (options.baseUrl) {
const hasScheme = /^\w[\w\d+.-]*:/.test(href);
if (!hasScheme) {
href = resolvePath(options.baseUrl, href).toString();
}
}
title = removeMarkdownEscapes(title);
href = removeMarkdownEscapes(href);
if (
!href
|| href.match(/^data:|javascript:/i)
|| (href.match(/^command:/i) && !markdown.isTrusted)
|| href.match(/^command:(\/\/\/)?_workbench\.downloadResource/i)
) {
// drop the link
return text;
} else {
// HTML Encode href
href = href.replace(/&/g, '&amp;')
.replace(/</g, '&lt;')
.replace(/>/g, '&gt;')
.replace(/"/g, '&quot;')
.replace(/'/g, '&#39;');
return `<a href="#" data-href="${href}" title="${title || href}">${text}</a>`;
}
};
renderer.paragraph = (text): string => {
if (markdown.supportThemeIcons) {
const elements = renderLabelWithIcons(text);
text = elements.map(e => typeof e === 'string' ? e : e.outerHTML).join('');
}
return `<p>${text}</p>`;
};
if (options.codeBlockRenderer) {
renderer.code = (code, lang) => {
const value = options.codeBlockRenderer!(lang, code);
// when code-block rendering is async we return sync
// but update the node with the real result later.
const id = defaultGenerator.nextId();
const promise = Promise.all([value, withInnerHTML]).then(values => {
const span = <HTMLDivElement>element.querySelector(`div[data-code="${id}"]`);
if (span) {
DOM.reset(span, values[0]);
}
}).catch(_err => {
// ignore
});
if (options.asyncRenderCallback) {
promise.then(options.asyncRenderCallback);
}
return `<div class="code" data-code="${id}">${escape(code)}</div>`;
};
}
if (options.actionHandler) {
const onClick = options.actionHandler.disposables.add(new DomEmitter(element, 'click'));
const onAuxClick = options.actionHandler.disposables.add(new DomEmitter(element, 'auxclick'));
options.actionHandler.disposables.add(Event.any(onClick.event, onAuxClick.event)(e => {
const mouseEvent = new StandardMouseEvent(e);
if (!mouseEvent.leftButton && !mouseEvent.middleButton) {
return;
}
let target: HTMLElement | null = mouseEvent.target;
if (target.tagName !== 'A') {
target = target.parentElement;
if (!target || target.tagName !== 'A') {
return;
}
}
try {
const href = target.dataset['href'];
if (href) {
options.actionHandler!.callback(href, mouseEvent);
}
} catch (err) {
onUnexpectedError(err);
} finally {
mouseEvent.preventDefault();
}
}));
}
// Use our own sanitizer so that we can let through only spans.
// Otherwise, we'd be letting all html be rendered.
// If we want to allow markdown permitted tags, then we can delete sanitizer and sanitize.
// We always pass the output through dompurify after this so that we don't rely on
// marked for sanitization.
markedOptions.sanitizer = (html: string): string => {
const match = markdown.isTrusted ? html.match(/^(<span[^>]+>)|(<\/\s*span>)$/) : undefined;
return match ? html : '';
};
markedOptions.sanitize = true;
markedOptions.silent = true;
markedOptions.renderer = renderer;
// values that are too long will freeze the UI
let value = markdown.value ?? '';
if (value.length > 100_000) {
value = `${value.substr(0, 100_000)}`;
}
// escape theme icons
if (markdown.supportThemeIcons) {
value = markdownEscapeEscapedIcons(value);
}
const renderedMarkdown = marked.parse(value, markedOptions);
element.innerHTML = sanitizeRenderedMarkdown(markdown, renderedMarkdown) as unknown as string;
// signal that async code blocks can be now be inserted
signalInnerHTML!();
// signal size changes for image tags
if (options.asyncRenderCallback) {
for (const img of element.getElementsByTagName('img')) {
const listener = DOM.addDisposableListener(img, 'load', () => {
listener.dispose();
options.asyncRenderCallback!();
});
}
}
return element;
}
function sanitizeRenderedMarkdown(
options: { isTrusted?: boolean },
renderedMarkdown: string,
): TrustedHTML {
const { config, allowedSchemes } = getSanitizerOptions(options);
dompurify.addHook('uponSanitizeAttribute', (element, e) => {
if (e.attrName === 'style' || e.attrName === 'class') {
if (element.tagName === 'SPAN') {
if (e.attrName === 'style') {
e.keepAttr = /^(color\:#[0-9a-fA-F]+;)?(background-color\:#[0-9a-fA-F]+;)?$/.test(e.attrValue);
return;
} else if (e.attrName === 'class') {
e.keepAttr = /^codicon codicon-[a-z\-]+( codicon-modifier-[a-z\-]+)?$/.test(e.attrValue);
return;
}
}
e.keepAttr = false;
return;
}
});
// build an anchor to map URLs to
const anchor = document.createElement('a');
// https://github.com/cure53/DOMPurify/blob/main/demos/hooks-scheme-allowlist.html
dompurify.addHook('afterSanitizeAttributes', (node) => {
// check all href/src attributes for validity
for (const attr of ['href', 'src']) {
if (node.hasAttribute(attr)) {
anchor.href = node.getAttribute(attr) as string;
if (!allowedSchemes.includes(anchor.protocol.replace(/:$/, ''))) {
node.removeAttribute(attr);
}
}
}
});
try {
return dompurify.sanitize(renderedMarkdown, { ...config, RETURN_TRUSTED_TYPE: true });
} finally {
dompurify.removeHook('uponSanitizeAttribute');
dompurify.removeHook('afterSanitizeAttributes');
}
}
function getSanitizerOptions(options: { readonly isTrusted?: boolean }): { config: dompurify.Config, allowedSchemes: string[] } {
const allowedSchemes = [
Schemas.http,
Schemas.https,
Schemas.mailto,
Schemas.data,
Schemas.file,
Schemas.vscodeFileResource,
Schemas.vscodeRemote,
Schemas.vscodeRemoteResource,
];
if (options.isTrusted) {
allowedSchemes.push(Schemas.command);
}
return {
config: {
// allowedTags should included everything that markdown renders to.
// Since we have our own sanitize function for marked, it's possible we missed some tag so let dompurify make sure.
// HTML tags that can result from markdown are from reading https://spec.commonmark.org/0.29/
// HTML table tags that can result from markdown are from https://github.github.com/gfm/#tables-extension-
ALLOWED_TAGS: ['ul', 'li', 'p', 'b', 'i', 'code', 'blockquote', 'ol', 'h1', 'h2', 'h3', 'h4', 'h5', 'h6', 'hr', 'em', 'pre', 'table', 'thead', 'tbody', 'tr', 'th', 'td', 'div', 'del', 'a', 'strong', 'br', 'img', 'span'],
ALLOWED_ATTR: ['href', 'data-href', 'target', 'title', 'src', 'alt', 'class', 'style', 'data-code', 'width', 'height', 'align'],
ALLOW_UNKNOWN_PROTOCOLS: true,
},
allowedSchemes
};
}
/**
* Strips all markdown from `string`, if it's an IMarkdownString. For example
* `# Header` would be output as `Header`. If it's not, the string is returned.
*/
export function renderStringAsPlaintext(string: IMarkdownString | string) {
return typeof string === 'string' ? string : renderMarkdownAsPlaintext(string);
}
/**
* Strips all markdown from `markdown`. For example `# Header` would be output as `Header`.
*/
export function renderMarkdownAsPlaintext(markdown: IMarkdownString) {
const renderer = new marked.Renderer();
renderer.code = (code: string): string => {
return code;
};
renderer.blockquote = (quote: string): string => {
return quote;
};
renderer.html = (_html: string): string => {
return '';
};
renderer.heading = (text: string, _level: 1 | 2 | 3 | 4 | 5 | 6, _raw: string): string => {
return text + '\n';
};
renderer.hr = (): string => {
return '';
};
renderer.list = (body: string, _ordered: boolean): string => {
return body;
};
renderer.listitem = (text: string): string => {
return text + '\n';
};
renderer.paragraph = (text: string): string => {
return text + '\n';
};
renderer.table = (header: string, body: string): string => {
return header + body + '\n';
};
renderer.tablerow = (content: string): string => {
return content;
};
renderer.tablecell = (content: string, _flags: {
header: boolean;
align: 'center' | 'left' | 'right' | null;
}): string => {
return content + ' ';
};
renderer.strong = (text: string): string => {
return text;
};
renderer.em = (text: string): string => {
return text;
};
renderer.codespan = (code: string): string => {
return code;
};
renderer.br = (): string => {
return '\n';
};
renderer.del = (text: string): string => {
return text;
};
renderer.image = (_href: string, _title: string, _text: string): string => {
return '';
};
renderer.text = (text: string): string => {
return text;
};
renderer.link = (_href: string, _title: string, text: string): string => {
return text;
};
// values that are too long will freeze the UI
let value = markdown.value ?? '';
if (value.length > 100_000) {
value = `${value.substr(0, 100_000)}`;
}
const unescapeInfo = new Map<string, string>([
['&quot;', '"'],
['&nbsp;', ' '],
['&amp;', '&'],
['&#39;', '\''],
['&lt;', '<'],
['&gt;', '>'],
]);
const html = marked.parse(value, { renderer }).replace(/&(#\d+|[a-zA-Z]+);/g, m => unescapeInfo.get(m) ?? m);
return sanitizeRenderedMarkdown({ isTrusted: false }, html).toString();
}