Open platform to collect and prioritize feedback
fix(markdown): escape code, alt, title and href under marked 17
Matt Robertsa1f5ba01a5f5
diff --git a/public/services/markdown.spec.ts b/public/services/markdown.spec.tsindex dd4398f4e..b87d2c01d 100644--- a/public/services/markdown.spec.ts+++ b/public/services/markdown.spec.ts@@ -32,7 +32,8 @@ const testCases = [},{input: `[Uh oh...]("onerror="alert('XSS'))`,- expectedFull: '<p><a class="text-link" href="" rel="noopener nofollow" target="_blank">Uh oh...</a></p>',+ // The quotes stay inside the (harmless, relative) href instead of breaking out of it+ expectedFull: `<p><a class="text-link" href=""onerror="alert('XSS')" rel="noopener nofollow" target="_blank">Uh oh...</a></p>`,expectedPlainText: "Uh oh...",},{@@ -87,6 +88,13 @@ testCases.forEach((x) => {})})+// Parse rendered HTML the way the browser would and list the live elements it creates.+const liveElements = (html: string): Element[] => {+ const container = document.createElement("div")+ container.innerHTML = html+ return Array.from(container.querySelectorAll("*"))+}+describe("XSS prevention", () => {const xssInputs = ["<script>alert(1)</script>",@@ -97,6 +105,9 @@ describe("XSS prevention", () => {"<img src=x onerror=alert(1)>","<script>alert(1)</script>","`<img src=x onerror=alert(1)>`",+ // Code spans/blocks are not escaped by marked 17's lexer; the renderers must escape them+ "`<img src=x onerror=alert(1)>`",+ "```\n<img src=x onerror=alert(1)>\n```",]xssInputs.forEach((input) => {@@ -118,3 +129,84 @@ describe("XSS prevention", () => {})})})++describe("code is rendered as text", () => {+ const inputs = ["`<b>`", "`<b>`", "`<img src=x onerror=alert(1)>`", "`<img src=x onerror=alert(1)>`", "```\n<b>bold</b>\n```"]++ inputs.forEach((input) => {+ test(`full mode creates no markup from code: ${input}`, () => {+ const tags = liveElements(markdown.full(input)).map((e) => e.tagName.toLowerCase())+ expect(tags.filter((t) => t !== "p" && t !== "code" && t !== "pre")).toEqual([])+ })++ test(`plainText mode creates no markup from code: ${input}`, () => {+ expect(liveElements(markdown.plainText(input))).toEqual([])+ })+ })+})++describe("plainText returns HTML-safe text", () => {+ test("does not decode entities back into markup", () => {+ expect(liveElements(markdown.plainText("Hello <b>Beautiful</b> World"))).toEqual([])+ expect(liveElements(markdown.plainText("<b>x</b>"))).toEqual([])+ })++ test("displays special characters as typed", () => {+ const container = document.createElement("p")+ container.innerHTML = markdown.plainText("Jane's & Jim's > [Matt](https://example.com) <b>hi</b>")+ expect(container.textContent).toEqual("Jane's & Jim's > Matt <b>hi</b>")+ })+})++describe("attribute values are escaped", () => {+ test("fider-image alt text cannot inject attributes", () => {+ const [p, img] = liveElements(markdown.full(''))+ expect(p.tagName).toEqual("P")+ expect(img.tagName).toEqual("IMG")+ expect(img.getAttribute("class")).toEqual("fider-inline-image")+ expect(img.getAttribute("alt")).toEqual('x" class="evil" data-x="1')+ expect(img.hasAttribute("data-x")).toBe(false)+ })++ test("link title is quoted", () => {+ const [, a] = liveElements(markdown.full('[a](http://x.com "hello world")'))+ expect(a.getAttribute("title")).toEqual("hello world")+ expect(a.hasAttribute("world")).toBe(false)+ })++ test("link title quotes cannot inject attributes", () => {+ const [, a] = liveElements(markdown.full(`[a](http://x.com 'say "hi" style="color:red"')`))+ expect(a.getAttribute("title")).toEqual('say "hi" style="color:red"')+ expect(a.hasAttribute("style")).toBe(false)+ })++ test("link href quotes cannot inject attributes", () => {+ const [, a] = liveElements(markdown.full('[a](http://x.com/"style="color:red)'))+ expect(a.tagName).toEqual("A")+ expect(a.hasAttribute("style")).toBe(false)+ })+})++describe("fider-image references", () => {+ const unsafe = [+ "",+ "",+ "",+ "",+ ]++ unsafe.forEach((input) => {+ test(`is not rendered as an inline image: ${input}`, () => {+ const html = markdown.full(input)+ expect(html).not.toContain("/static/images/")+ expect(html).not.toContain("data-bkey")+ })+ })+})++describe("mentions", () => {+ test("renders non-empty @[name] as a mention and leaves @[] as text", () => {+ const spans = liveElements(markdown.full("@[] and @[Jane Doe]")).filter((e) => e.classList.contains("mention"))+ expect(spans.map((e) => e.textContent)).toEqual(["@Jane Doe"])+ })+})diff --git a/public/services/markdown.ts b/public/services/markdown.tsindex 600807a3d..80a37dae6 100644--- a/public/services/markdown.ts+++ b/public/services/markdown.ts@@ -1,6 +1,7 @@import { Marked, RendererObject } from "marked"import DOMPurify from "dompurify"import { fiderAllowedSchemes } from "@fider/hooks"+import { fiderImageBkey, MENTION_GLOBAL } from "./markdownSyntax"if (DOMPurify.isSupported) {DOMPurify.setConfig({@@ -26,8 +27,12 @@ if (DOMPurify.isSupported) {})}-// marked 5+ escapes text (including ' -> '). We keep apostrophes literal to match the-// historical output, while preserving existing entities and escaping & < > ".+// marked 5+ no longer escapes token text in the lexer; escaping is the renderer's job, so every+// override that writes token text/attributes into HTML must escape it itself.++// Text and attribute values. marked 5+ escapes text (including ' -> '); we keep apostrophes+// literal to match the historical output, while preserving existing entities (input has already+// been through encodeHTML, so "<" arrives as "<") and escaping & < > ".const escapeText = (s: string): string =>s.replace(/&(?![#\w]+;)/g, "&")@@ -35,8 +40,12 @@ const escapeText = (s: string): string =>.replace(/>/g, ">").replace(/"/g, """)+// Code (spans and blocks): escape everything, exactly like marked's default code renderers,+// so plainText and full show code the same way.+const escapeCode = (s: string): string => s.replace(/&/g, "&").replace(/</g, "<").replace(/>/g, ">").replace(/"/g, """).replace(/'/g, "'")+// Expand Fider's proprietary mention syntax @[name] into a styled span.-const renderMentions = (html: string): string => html.replace(/@\[(.*?)\]/g, (_match, name) => `<span class="mention">@${name}</span>`)+const renderMentions = (html: string): string => html.replace(MENTION_GLOBAL, (_match, name) => `<span class="mention">@${name}</span>`)// marked 5+ replaced the positional Renderer API with token objects and requires renderer// overrides to be registered via .use(). Typing as RendererObject binds `this` to the@@ -44,17 +53,18 @@ const renderMentions = (html: string): string => html.replace(/@\[(.*?)\]/g, (_m// keep marked's defaults.const fullRenderer: RendererObject = {image({ href, text }) {- // Fider's proprietary inline-image syntax: - if (href && href.startsWith("fider-image:")) {- const bkey = href.substring("fider-image:".length)- return `<img src="/static/images/${bkey}" alt="${text || ""}" class="fider-inline-image" data-bkey="${bkey}" />`+ // Fider's proprietary inline-image syntax: . The bkey is validated+ // (safe characters, no "..") so it can go into the attributes as-is.+ const bkey = fiderImageBkey(href)+ if (bkey) {+ return `<img src="/static/images/${bkey}" alt="${escapeText(text || "")}" class="fider-inline-image" data-bkey="${bkey}" />`}return false // fall back to marked's default image renderer},link({ href, title, tokens }) {const text = this.parser.parseInline(tokens)- const titleAttr = title ? ` title=${title}` : ""- return `<a class="text-link" href="${href}"${titleAttr} rel="noopener nofollow" target="_blank">${text}</a>`+ const titleAttr = title ? ` title="${escapeText(title)}"` : ""+ return `<a class="text-link" href="${escapeText(href)}"${titleAttr} rel="noopener nofollow" target="_blank">${text}</a>`},text(token) {const rendered = "tokens" in token && token.tokens ? this.parser.parseInline(token.tokens) : escapeText(token.text)@@ -88,13 +98,13 @@ const plainTextRenderer: RendererObject = {return ` ${this.parser.parseInline(tokens)} `},code({ text }) {- return text+ return escapeCode(text)},codespan({ text }) {- return text+ return escapeCode(text)},html({ text }) {- return text+ return escapeText(text)},blockquote({ tokens }) {return this.parser.parse(tokens)diff --git a/public/services/markdownSyntax.ts b/public/services/markdownSyntax.tsnew file mode 100644index 000000000..89dee40be--- /dev/null+++ b/public/services/markdownSyntax.ts@@ -0,0 +1,26 @@+// Fider's proprietary markdown syntax. Shared by the display renderer (services/markdown.ts)+// and the comment editor (CustomMention / CustomImage) so both read stored content the same way.++// Mentions: @[Display Name]. The name is non-empty, single-line and ends at the first "]"+// (so "@[]" is plain text and never swallows a later "]").+const MENTION_SOURCE = "@\\[([^\\]\\n]+)\\]"+export const MENTION_START = "@["+// Anchored: matches a mention at the start of the input (editor tokenizer).+export const MENTION_AT_START = new RegExp(`^${MENTION_SOURCE}`)+// Global: finds every mention in a string (display renderer). String.replace resets lastIndex.+export const MENTION_GLOBAL = new RegExp(MENTION_SOURCE, "g")++// Inline images: . Only bkeys in the shape the server/editor generate+// (e.g. attachments/<random>-<name>.png) are accepted, and never with ".." path traversal,+// because the bkey is appended to /static/images/.+const FIDER_IMAGE_PREFIX = "fider-image:"+const BKEY_PATTERN = /^[a-zA-Z0-9_/.-]+$/++export const fiderImageBkey = (href: string | null | undefined): string | undefined => {+ if (!href || !href.startsWith(FIDER_IMAGE_PREFIX)) return undefined+ const bkey = href.substring(FIDER_IMAGE_PREFIX.length)+ if (!BKEY_PATTERN.test(bkey) || bkey.includes("..")) return undefined+ return bkey+}++export const fiderImageMarkdown = (bkey: string): string => ``