Alexsey’s TTPs

(.. Tactics, Techniques, and Procedures)

Mike Arpaia and I flew out of JFK on the morning of January 22nd, 2012. We were gainfully employed by ████████ and responding to a breach that a client had suffered. Mike and I spent a week pouring over forensic artifacts and soon identified the perp as a Russian-speaking hacker called “M4g”.

M4g would later be revealed by the FBI as Alexsey Belan, indicted four times (including the Yahoo hack), and sanctioned by the Obama administration.
Photos of Alexsey Belan published by the FBI

Belan targeted tech firms on the west coast. Many of the large breaches publicized during 2012 and 2013 are attributed to him, and news of others didn’t make it into the public domain.

This post details the victim estates and provides insight into the modus operandi of a prolific adversary. Please use this material to consider your environment and ensure these weak links do not exist within it.

TL;DR

Belan’s observed offensive traits were as follows:

  • He identified peripheral web servers via Google and Linkedin searches
  • Used known WordPress flaws and custom bugs to compromise PHP sites
  • Linux authentication mechanisms were altered to capture credentials
  • Nmap was used to identify exposed network services internally
  • Corporate Wikis revealed administrative workflows and VPN details
  • Ticketing, bug tracking, and version control systems provided secrets (e.g. cryptographic keys, seeds, hashes, credentials, and source code)
  • Cookies from weak non-production instances (e.g. staging) were valid in production as cryptographic materials were the same — bypassing 2FA
  • Client certificates (exposed by email, ticketing, or lifted from filesystems) were combined with known credentials to access corporate VPNs
  • Engineering credentials were used to commit backdoors to version control which were self-approved and later deployed into production

Email addresses and password hashes were amassed with each compromise. Cracked credentials were used to target further victims via exposed mail services (e.g. Outlook on the Web or G Suite), and the exploratory process was repeated to gain privileged network access via VPN or similar means.

Defensive controls checklist:

  • Segregate risky Internet-exposed servers (e.g. third-party PHP sites)
  • Do not leave secrets in your Wiki, ticketing, or bug tracking systems, or send sensitive material over email (e.g. credentials, certificates, or keys)
  • Don’t re-use cryptographic seeds, keys, or credentials across production and non-production environments (e.g. development or staging)
  • Compromised servers made odd DNS requests (e.g. resolving m4g.ru). Use a monitoring tool such as DNS Analytics for Splunk to flag anomalies
  • Software certificates are an insufficient second factor. Use a real 2FA solution such as Duo and consider hardware tokens to protect private keys via FIPS 201–1 and other standards, as supported by the YubiKey 4
Proper segmentation, enforcement of a sensible policy around the safe transmission of secrets, and close monitoring of production systems nullify this threat.

Technical Details

The marketing department of Belan’s first known victim ran a vulnerable WordPress server from a branch office. The server sat on the corporate WAN and could access internal systems: it wasn’t segregated or placed in a DMZ.

Belan found the server through a Google search and exploited it via CVE-2011–4106. Next, he downloaded tools via wget from m4g.ru and elevated privileges via CVE-2010–3856.

The following steps were undertaken over a five day period:

  • He replaced the PHP authentication mechanism (l10n.php) to capture credentials of corporate users interacting with the WordPress site
  • Ran Nmap and discovered the internal Wiki
  • Captured a handful of valid credentials via WordPress
  • Authenticated with the Wiki using captured user credentials
  • Came to understand the administrative procedures via the Wiki
  • Failed to authenticate with the administrative portal (2FA was required)
  • Identified a staging instance of the administrative portal that didn’t require 2FA. Authenticated, and subsequently identified directory traversal and arbitrary file upload flaws through manual testing
  • Found that cookies generated in staging were valid in production, and modified a staging cookie to access the production instance of the administrative portal (bypassing 2FA)
  • Gained privileged access to the administrative portal server via the known web application flaws, and a obtained a MySQL credential that provided access to customer data within the production environment
  • Ran mysqldump against the production database, creating 1.txt
  • Compressed and attempted to SCP 1.tgz to a masterhost.ru shell server using directory structures containing mnogo.mobi and m4g.ru domains
  • The SCP process failed, and so 1.tgz was split into chunks and exfiltrated by placing them onto an NFS share used to serve static content in the production web environment

Use of Linkedin to Target Peripheral Systems

Belan’s second known victim was notified by the FBI of a breach. We identified an engineering account that had accessed the corporate VPN from Russian IP space, and subsequently discovered a compromised web server running on the engineer’s iMac in the kitchen of his home in Santa Clara.

The web server was a Linux VM running within the iMac via Parallels. Belan had found the server via the engineer’s own public Linkedin profile, e.g.

Personal websites listed in a user’s Linkedin profile

The Linux VM hosted a handful of PHP sites. Belan worked to remotely compromise the Linux VM, host system, and finally his target, as follows:

  • Tested each PHP site by hand to identify potential weaknesses
  • Successfully identified and exploited a custom arbitrary file upload flaw
  • Elevated privileges to root via CVE-2010–3856 (as before)
  • Altered Linux and PHP authentication mechanisms to capture credentials
  • Obtained and cracked local user password hashes from /etc/shadow
  • Authenticated with the Linux VM via SSH using a valid credential
  • Launched a brute-force SSH password grinding attack from the guest Linux VM against the host operating system (running MacOS X)
  • Used tools to clear utmp and wtmp log entries within the Linux VM
  • Successfully authenticated with the host operating system via SSH using a permutation of a known user credential
  • Obtained corporate VPN configuration details from the host operating system, including the gateway details, client certificate, and private key
  • Combined the VPN settings with known credentials to authenticate with, and access the corporate VPN as the engineer himself

Application, database, and infrastructure logs within the victim environment were rolling (overwriting entries after 10–14 days). As such, our ability to investigate this compromise was limited. The facts however are as follows:

  • Belan maintained corporate VPN access for four months
  • The customer database was obtained from the production environment

Lack of 2FA + The Cloud =

By mid-2013, Belan had amassed 200 million credentials (including email addresses, passwords, but also answers to security questions). Organizations embracing cloud services but not 2FA were soft targets.

A third victim was breached through the following steps:

  • Use of a valid credential (user/pass) to authenticate with Google Mail
  • Access to an Internet-based JIRA instance granted via OpenID
  • Issues in JIRA revealed a legacy Internet-based Subversion (SVN) server
  • Already known and valid credentials were used to access the SVN server
  • Local privileges were elevated and engineering password hashes were obtained from /etc/shadow and other files within version control
  • Internet-exposed Git production instance accessed via cracked credentials
  • JSP shell committed into Git, self-approved, and queued for deployment
  • An unwitting engineer later deployed the code into production
  • Belan read the production database environment variables from an application server via the JSP shell he had implanted
  • Ran mysqldump against the production database, creating 1.txt
  • Compressed, split, and exfiltrated 1.tgz via the application server

Fallout and Closing Remarks

1.2 billion usernames, password hashes, and security questions have been compromised from a handful of known victims (including Yahoo, Evernote, Scribd, and Zappos, according to the New York Times), and likely millions of further records from unknown victims.

Consider the number of organizations that provide services to their users and employees over the public Internet, including:

  • Web portals for sales and marketing purposes
  • Mail access via Microsoft Outlook on the Web and Google Mail
  • Collaboration via Slack, HipChat, SharePoint, and Confluence
  • DevOps and support via GitHub, JIRA, and CI/CD utilities

Next, consider how many enforce 2FA across their entire attack surface. Large enterprises often expose domain-joined systems to the Internet that can be leveraged to provide privileged network access (via Microsoft IIS, SharePoint, and other services supporting NTLM authentication).

The number of weak networks is high enough for Belan and the FSB to pose a serious threat to organizations around the globe with their Rolodex of secrets. The personal mail and messaging accounts of politicians, lawyers, activists, and journalists can also be easily targeted.

Finally, consider the number of corporate VPNs using unsafe 2FA (requiring only a username, password, and software-based certificate). First-hand experience leads me to believe this number is high. VPN certificates and keys are often found within and lifted from email, ticketing, and chat services. 💥