TL;DR
Pluck CMS 4.7.20 protects its admin panel against cross-site request forgery with a function that checks whether a request came from the same domain, by reading the Referer header. The catch: when there is no Referer at all, the check logs a warning and then returns true anyway (CVE-2026-78748). So an attacker's page that simply suppresses the Referer can make a logged-in admin's browser fire off any admin action, from editing pages to installing modules, with the admin's own session. Block8.ai's Security Research Team rates it CVSS 3.1 of 8.8 (High), CWE-352.
What the admin panel is trying to guard against
Cross-site request forgery is the trick where your logged-in session gets borrowed. You are signed into an admin panel in one tab. In another tab you open a page that quietly submits a form to that admin panel. Your browser attaches your session cookie automatically, because that is what browsers do, and the admin panel sees a perfectly authenticated request it never meant to receive.
The standard defence is to confirm the request genuinely originated from the application itself. Pluck's approach is to look at the Referer header, the browser's note about which page the request came from, and check it points back at the site's own domain.
That is a reasonable first line, as long as you handle the awkward case: what do you do when the Referer is simply not there?
Where the trust breaks
Pluck answers that question in the worst possible way. The relevant branch in requestedByTheSameDomain(), in data/inc/functions.admin.php, reads roughly:
elseif ($myDomain == NULL || $requestsSource == NULL) {
show_error("Be carefull...");
return true;
}
When the Referer is absent, $requestsSource is NULL. The function notices, shows a warning, and then returns true. In other words, it treats "I cannot tell where this came from" as "this is fine". Every caller that trusts this function to mean "the request is legitimate" is now trusting a function that says yes to the unknown.
And an attacker can make the Referer absent. A page served with Referrer-Policy: no-referrer, or a form element carrying referrerpolicy="no-referrer", instructs the browser to omit the header entirely. In several configurations browsers drop it by default anyway. So the attacker isn't working around the check; they are walking through the gap the check leaves open on purpose.
Under the bonnet
Across admin.php, every sensitive action is gated on a CSRF variable derived from this function: installmodule, themeinstall, editpage, settings, deleteimage, deletefile, deletepage, and more. One flawed function sits upstream of all of them.
The attack page is unremarkable:
<form method="post"
action="http://target/admin.php?action=installmodule"
referrerpolicy="no-referrer"
enctype="multipart/form-data">
<input type="file" name="sendfile" value="malicious_shell.tar.gz.php" />
<input type="submit" name="submit" value="Upload" />
</form>
When a logged-in admin lands on this page and the form submits, the browser omits the Referer, requestedByTheSameDomain() prints its "Be carefull..." warning, returns true, and the admin action runs under the victim's session. In testing, sending admin.php?action=installmodule with no Referer executed the module install path, confirmed by a PharData exception that only occurs after the uploaded file has already been written to data/modules/.
This is CWE-352 in its textbook form: a state-changing action with no unguessable, request-bound token, relying instead on a header the attacker controls.
Why it matters
The severity here comes from scope. This is not one vulnerable action; it is every admin action in the panel. An attacker who can get an admin to open a single web page can forge page edits, delete files, change settings, or install components, all without the admin clicking anything meaningful.
The 8.8 rating reflects the worst reachable outcome. The installmodule action, chained with the module-install filename bypass tracked separately, turns a forged request into a planted PHP webshell and from there into full remote code execution. The Referer-based check was the one thing standing between "a malicious link" and "a compromised server", and it holds the door open for anyone who arrives without a name tag.
It is worth noting this is close kin to CVE-2020-24740, which covered CSRF on the editpage action in older Pluck releases. Same root cause, different version, and here with the net cast across the entire admin surface.
How to fix it
- Use per-request anti-CSRF tokens. Generate an unpredictable token tied to the session, embed it in every admin form, and verify it server-side on every state-changing request. This does not depend on headers the attacker can strip.
- Fail closed. If the function cannot determine the request's origin, it must return
false, nottrue. A missingReferershould deny the action, never approve it. - Treat the
Refereras a secondary signal at best. It is optional, spoofable, and frequently absent for legitimate privacy reasons, so it can never be the sole gate. - Consider the
Originheader and theSameSitecookie attribute as additional layers, but tokens remain the primary defence.
Until a fixed release ships, administrators should avoid staying logged in while browsing untrusted pages, and operators should restrict access to the admin panel where they can.
Closing
A security check that returns "yes" when it has no idea is worse than no check at all, because it buys false confidence. Pluck's function did exactly what it was asked, noticed it couldn't verify the origin, announced the fact, and approved the request regardless. The Block8.ai Security Research Team found this during authorised research into Pluck CMS. When you build a gate, decide what happens on uncertainty before anything else, and make sure the answer is "stop".