joplin/packages/renderer/htmlUtils.ts

const Entities = require('html-entities').AllHtmlEntities;
const htmlentities = new Entities().encode;
import { fileUriToPath } from '@joplin/utils/url';
const htmlparser2 = require('@joplin/fork-htmlparser2');

// [\s\S] instead of . for multiline matching
// https://stackoverflow.com/a/16119722/561309
const imageRegex = /<img([\s\S]*?)src=["']([\s\S]*?)["']([\s\S]*?)>/gi;

const anchorRegex = /<a([\s\S]*?)href=["']([\s\S]*?)["']([\s\S]*?)>/gi;

const selfClosingElements = [
	'area',
	'base',
	'basefont',
	'br',
	'col',
	'command',
	'embed',
	'frame',
	'hr',
	'img',
	'input',
	'isindex',
	'keygen',
	'link',
	'meta',
	'param',
	'source',
	'track',
	'wbr',
];

interface SanitizeHtmlOptions {
	addNoMdConvClass?: boolean;
	allowedFilePrefixes?: string[];
}

export const attributesHtml = (attr: Record<string, string>) => {
	const output = [];

	for (const n in attr) {
		if (!attr.hasOwnProperty(n)) continue;

		if (!attr[n]) {
			output.push(n);
		} else {
			output.push(`${n}="${htmlentities(attr[n])}"`);
		}
	}

	return output.join(' ');
};

export const isSelfClosingTag = (tagName: string) => {
	return selfClosingElements.includes(tagName.toLowerCase());
};

class HtmlUtils {

	// eslint-disable-next-line @typescript-eslint/ban-types -- Old code before rule was applied
	public processImageTags(html: string, callback: Function) {
		if (!html) return '';

		return html.replace(imageRegex, (_v, before, src, after) => {
			const action = callback({ src: src });

			if (!action) return `<img${before}src="${src}"${after}>`;

			if (action.type === 'replaceElement') {
				return action.html;
			}

			if (action.type === 'replaceSource') {
				return `<img${before}src="${action.src}"${after}>`;
			}

			if (action.type === 'setAttributes') {
				const attrHtml = attributesHtml(action.attrs);
				return `<img${before}${attrHtml}${after}>`;
			}

			throw new Error(`Invalid action: ${action.type}`);
		});
	}

	// eslint-disable-next-line @typescript-eslint/ban-types -- Old code before rule was applied
	public processAnchorTags(html: string, callback: Function) {
		if (!html) return '';

		interface Action {
			type: 'replaceElement' | 'replaceSource' | 'setAttributes';
			href: string;
			html: string;
			attrs: Record<string, string>;
		}

		return html.replace(anchorRegex, (_v, before, href, after) => {
			const action: Action = callback({ href: href });

			if (!action) return `<a${before}href="${href}"${after}>`;

			if (action.type === 'replaceElement') {
				return action.html;
			}

			if (action.type === 'replaceSource') {
				return `<img${before}href="${action.href}"${after}>`;
			}

			if (action.type === 'setAttributes') {
				const attrHtml = attributesHtml(action.attrs);
				return `<img${before}${attrHtml}${after}>`;
			}

			throw new Error(`Invalid action: ${action.type}`);
		});
	}

	public stripHtml(html: string) {
		const output: string[] = [];

		const tagStack: string[] = [];

		const currentTag = () => {
			if (!tagStack.length) return '';
			return tagStack[tagStack.length - 1];
		};

		const disallowedTags = ['script', 'style', 'head', 'iframe', 'frameset', 'frame', 'object', 'base'];

		const parser = new htmlparser2.Parser({

			onopentag: (name: string) => {
				tagStack.push(name.toLowerCase());
			},

			ontext: (decodedText: string) => {
				if (disallowedTags.includes(currentTag())) return;
				output.push(decodedText);
			},

			onclosetag: (name: string) => {
				if (currentTag() === name.toLowerCase()) tagStack.pop();
			},

		}, { decodeEntities: true });

		parser.write(html);
		parser.end();

		// In general, we want to get back plain text from this function, so all
		// HTML entities are decoded. Howver, to prevent XSS attacks, we
		// re-encode all the "<" characters, which should break any attempt to
		// inject HTML tags.

		return output.join('')
			.replace(/\s+/g, ' ')
			.replace(/</g, '&lt;');
	}

	// This is tested in sanitize_links.md
	private isAcceptedUrl(url: string, allowedFilePrefixes: string[]): boolean {
		url = url.toLowerCase();
		if (url.startsWith('https://') ||
			url.startsWith('http://') ||
			url.startsWith('mailto:') ||
			url.startsWith('joplin://') ||
			!!url.match(/:\/[0-9a-zA-Z]{32}/) ||
			// We also allow anchors but only with a specific set of a characters.
			// Fixes https://github.com/laurent22/joplin/issues/8286
			!!url.match(/^#[a-zA-Z0-9-]+$/)) return true;

		if (url.startsWith('file://')) {
			// We need to do a case insensitive comparison because the URL we
			// get appears to be converted to lowercase somewhere. To be
			// completely sure, we make it lowercase explicitely.
			const filePath = fileUriToPath(url).toLowerCase();
			for (const filePrefix of allowedFilePrefixes) {
				if (filePath.startsWith(filePrefix.toLowerCase())) return true;
			}
		}

		return false;
	}

	public sanitizeHtml(html: string, options: SanitizeHtmlOptions = null) {
		options = {
			// If true, adds a "jop-noMdConv" class to all the tags.
			// It can be used afterwards to restore HTML tags in Markdown.
			addNoMdConvClass: false,
			allowedFilePrefixes: [],
			...options,
		};

		const output: string[] = [];

		const tagStack: string[] = [];

		const currentTag = () => {
			if (!tagStack.length) return '';
			return tagStack[tagStack.length - 1];
		};

		// When we encounter a disallowed tag, all the other tags within it are
		// going to be skipped too. This is necessary to prevent certain XSS
		// attacks. See sanitize_11.md
		let disallowedTagDepth = 0;

		// The BASE tag allows changing the base URL from which files are
		// loaded, and that can break several plugins, such as Katex (which
		// needs to load CSS files using a relative URL). For that reason it is
		// disabled. More info: https://github.com/laurent22/joplin/issues/3021
		//
		// "link" can be used to escape the parser and inject JavaScript. Adding
		// "meta" too for the same reason as it shouldn't be used in notes
		// anyway.
		//
		// There are too many issues with SVG tags and to handle them properly
		// we should parse them separately. Currently we are not so it is better
		// to disable them. SVG graphics are still supported via the IMG tag.
		const disallowedTags = [
			'script', 'iframe', 'frameset', 'frame', 'object', 'base',
			'embed', 'link', 'meta', 'noscript', 'button', 'form',
			'input', 'select', 'textarea', 'option', 'optgroup',
			'svg',

			// Disallow map and area tags: <area ...> links are currently not
			// sanitized as well as <a ...> links, allowing potential sandbox
			// escape.
			'map', 'area',
		];

		const parser = new htmlparser2.Parser({

			onopentag: (name: string, attrs: Record<string, string>) => {
				// Note: "name" and attribute names are always lowercase even
				// when the input is not. So there is no need to call
				// "toLowerCase" on them.

				tagStack.push(name);

				if (disallowedTags.includes(currentTag())) {
					disallowedTagDepth++;
					return;
				}

				if (disallowedTagDepth) return;

				attrs = { ...attrs };

				// Remove all the attributes that start with "on", which
				// normally should be JavaScript events. A better solution
				// would be to blacklist known events only but it seems the
				// list is not well defined [0] and we don't want any to slip
				// throught the cracks. A side effect of this change is a
				// regular harmless attribute that starts with "on" will also
				// be removed.
				// 0: https://developer.mozilla.org/en-US/docs/Web/Events
				for (const attrName in attrs) {
					if (!attrs.hasOwnProperty(attrName)) continue;
					if (attrName.length <= 2) continue;
					if (attrName.substr(0, 2) !== 'on') continue;
					delete attrs[attrName];
				}

				// Make sure that only non-acceptable URLs are filtered out. In
				// particular we want to exclude `javascript:` URLs. This
				// applies to A tags, and also AREA ones but to be safe we don't
				// filter on the tag name and process all HREF attributes.
				if ('href' in attrs && !this.isAcceptedUrl(attrs['href'], options.allowedFilePrefixes)) {
					attrs['href'] = '#';
				}

				// We need to clear any such attribute, otherwise it will
				// make any arbitrary link open within the application.
				if ('data-from-md' in attrs) {
					delete attrs['data-from-md'];
				}

				if (options.addNoMdConvClass) {
					let classAttr = attrs['class'] || '';
					if (!classAttr.includes('jop-noMdConv')) {
						classAttr += ' jop-noMdConv';
						attrs['class'] = classAttr.trim();
					}
				}

				// For some reason, entire parts of HTML notes don't show up in
				// the viewer when there's an anchor tag without an "href"
				// attribute. It doesn't always happen and it seems to depend on
				// what else is in the note but in any case adding the "href"
				// fixes it. https://github.com/laurent22/joplin/issues/5687
				if (name === 'a' && !attrs['href']) {
					attrs['href'] = '#';
				}

				let attrHtml = attributesHtml(attrs);
				if (attrHtml) attrHtml = ` ${attrHtml}`;
				const closingSign = isSelfClosingTag(name) ? '/>' : '>';
				output.push(`<${name}${attrHtml}${closingSign}`);
			},

			ontext: (decodedText: string) => {
				if (disallowedTagDepth) return;


				if (currentTag() === 'style') {
					// For CSS, we have to put the style as-is inside the tag
					// because if we html-entities encode it, it's not going to
					// work. But it's ok because JavaScript won't run within the
					// style tag. Ideally CSS should be loaded from an external
					// file.

					// We however have to encode at least the `<` characters to
					// prevent certain XSS injections that would rely on the
					// content not being encoded (see sanitize_13.md)
					output.push(decodedText.replace(/</g, '&lt;'));
				} else {
					output.push(htmlentities(decodedText));
				}
			},

			onclosetag: (name: string) => {
				const current = currentTag();

				if (current === name.toLowerCase()) tagStack.pop();

				// The Markdown sanitization code can result in calls like this:
				//     sanitizeHtml('<invlaid>')
				//     sanitizeHtml('</invalid>')
				// Thus, we need to be able to remove '</invalid>', even if there is no
				// corresponding opening tag.
				if (disallowedTags.includes(current) || disallowedTags.includes(name)) {
					if (disallowedTagDepth > 0) {
						disallowedTagDepth--;
					}
					return;
				}

				if (disallowedTagDepth) return;

				if (isSelfClosingTag(name)) return;
				output.push(`</${name}>`);
			},

		}, { decodeEntities: true });

		parser.write(html);
		parser.end();

		return output.join('');
	}


}

const makeHtmlTag = (name: string, attrs: Record<string, string>) => {
	let attrHtml = attributesHtml(attrs);
	if (attrHtml) attrHtml = ` ${attrHtml}`;
	const closingSign = isSelfClosingTag(name) ? '/>' : '>';
	return `<${name}${attrHtml}${closingSign}`;
};

// Will return either the content of the <BODY> tag if it exists, or the whole
// HTML (which would be a fragment of HTML)
export const extractHtmlBody = (html: string) => {
	let inBody = false;
	let bodyFound = false;
	const output: string[] = [];

	const parser = new htmlparser2.Parser({

		onopentag: (name: string, attrs: Record<string, string>) => {
			if (name === 'body') {
				inBody = true;
				bodyFound = true;
				return;
			}

			if (inBody) {
				output.push(makeHtmlTag(name, attrs));
			}
		},

		ontext: (encodedText: string) => {
			if (inBody) output.push(encodedText);
		},

		onclosetag: (name: string) => {
			if (inBody && name === 'body') inBody = false;

			if (inBody) {
				if (isSelfClosingTag(name)) return;
				output.push(`</${name}>`);
			}
		},

	}, { decodeEntities: false });

	parser.write(html);
	parser.end();

	return bodyFound ? output.join('') : html;
};

export default new HtmlUtils();