Deception CheckBlog

Can your firewall reach ESXi over SSH?

In one intrusion, attackers took control of a Check Point gateway, used it to reach an ESXi host, and ran a wiper. Check both fixes, then check where the gateway itself can connect.

How the attackers reached the ESXi host

A hypervisor such as VMware ESXi runs virtual machines. SSH gives an authorized operator a way to open a command session on the host. In this case, Check Point Research traced an SSH login to a compromised gateway. The team reported that the attackers then ran a wiper, software designed to destroy data. Check Point Research investigation

  1. 01 · ENTRYMobile Access portalPHP2Shell runs code through the internet-facing web portal.
  2. 02 · CONTROLRoot on GaiaDynamicCLI flaw gives the attacker full operating-system privileges. Web shells and a backdoor follow.
  3. 03 · PIVOTGateway to ESXiAttackers scan internally, then log in over SSH using vpxuser.
  4. 04 · IMPACTWiper runsThe attackers execute destroy_v6 on the ESXi host.
Observed sequence in one intrusion documented by Check Point Research.

PHP2Shell (CVE-2026-109618 and CVE-2026-109619) was the way in. CVE-2026-123039 was the way to root: Gaia’s DynamicCLI service accepted commands from a local socket that any local user could write to. Check Point reports exploitation starting October 30, roughly a week after its PHP2Shell hotfix became available. CISA added the Gaia flaw to its Known Exploited Vulnerabilities catalog on November 13, marked it for forensic triage, and listed November 16, 2026 as the federal catalog due date. Check Point advisory · CISA KEV entry

Prioritize the chain. Check Point scored the Gaia flaw 7.8. It requires code running on the gateway, and PHP2Shell gave an outside attacker that foothold. When an internet-facing device has both flaws, treat them as one urgent remediation task and verify both fixes.

Verify both fixes on the devices you run

Prioritize internet-facing Mobile Access gateways. Include management systems in your patch plan and follow Check Point’s deployment order for your topology. In SmartConsole, check the installed Jumbo Hotfix Take in the Gateways & Servers view. Verify each applicable device or cluster member after the change. Check Point’s SmartConsole guidance

November 12 Jumbo Hotfix Takes. Each release note lists fixes for PHP2Shell and the Gaia escalation. Check Point lists these as the first fixed Takes for the Gaia flaw. Use the current supported Take for your branch that includes both fixes.
Gaia releaseVerified Take
R82.20Take 14
R82.10Take 52
R82Take 132
R81.20Take 174

Older enterprise releases. Check Point support article SK1000362 lists R81 and R81.10 as affected and gives fixed Takes for the four branches above. Standard support ended in October 2024 for enterprise R81 and in March 2026 for enterprise R81.10. Ask Check Point for a supported upgrade path and any applicable interim mitigation. Its workaround stops DynamicCLI and disables the commands it provides, so confirm the operational impact before applying it. Spark firewalls have separate PHP2Shell guidance. Check Point support article SK1000362 · Support lifecycle · Spark guidance

On an affected Gaia appliance where DynamicCLI is running, Check Point documents a read-only check: ls -l /tmp/dynamic_socket. The fixed socket shows mode 0660 and owner/group root config. Use that as a second check alongside the installed Take. Check Point verification guidance

Compare three records from the same window

Start with October 30 through November 4 for the activity Check Point describes. Extend the search through your own exposure window and use UTC when lining up records. Start with retained copies off the appliance when available.

1. Check the gateway

On each applicable gateway or cluster member, review /var/log/dynamicCli.log or its retained copy. Check Point recommends looking for Received = entries missing the normal Clish wrapper and entries containing Received = #return_value#. On patched systems, Rejected connection from unauthorized uid records a blocked call. Compare unexpected entries with portal and authentication activity around the same time. Check Point’s exact log checks

Check Mobile Access artifacts for PHP session files containing upload_progress_pwn, unexpected template changes, or a CPconfigdaemon binary under /tmp. If forwarded firewall logs have a gap, compare it with service restarts and collector health. Preserve relevant records and note the gateway and timestamp. Vendor indicators

2. Check the network path

Search independently collected firewall or flow records for traffic from the gateway’s own addresses to ESXi management addresses over SSH, typically TCP 22. Record source, destination, time, action, and any address translation. Check the source before translation or use session records to distinguish gateway-originated traffic from traffic forwarded through it. Compare any allowed flow with approved management routes and change records.

3. Check the ESXi host

Look for an accepted SSH login from the source seen by ESXi. Start with auth.log or its forwarded copy. On affected ESXi 8.0 Update 3 builds, also check syslog.log for sshd-session messages. Record the user, source, and timestamp. Use the World ID from the corresponding SSH session-open record to find commands in shell.log, where retained. vpxuser is a normal vCenter service account. The signal is an accepted SSH login from an unexpected source, regardless of username. Broadcom shell auditing · ESXi 8.0 Update 3 log location · vCenter service account

Read the results in order. An allow event shows traffic was permitted at that point. A completed TCP handshake can confirm a connection. An accepted SSH event shows authentication succeeded. The matching shell record can show what ran. If an unexpected login lines up with gateway compromise evidence, preserve the logs and escalate both systems to incident response. A missing match in one log leaves that step unresolved.

Check the route before you need it

Use the same check with any firewall or VPN gateway that can reach a virtualization management network. Pick one gateway and one host. Ask the network and virtualization owners to answer these five questions together.

  1. Is SSH running on the host? On ESXi, open the host’s Configure → System → Services view in vSphere Client. Review the service state and startup policy. If SSH is unnecessary, stop it and prevent automatic startup through change control after verifying another administrative path. ESXi Shell and SSH startup policy · Current vSphere Client menu path
  2. Can the gateway itself reach management over SSH? Check the rule, record which sources need access and who owns the exception, and remove unused access through change control. Run an authorized test from a representative source and find the corresponding record.
  3. Who patches the exposed portal? Identify its owner and confirm how they verify an emergency fix on each applicable member.
  4. Where do the records go? Ask one owner to find the network flow and another to find the host result. Make sure you can retrieve those records if the gateway is unavailable.
  5. Could the team restore a virtual machine if the host were lost? Verify that backups can survive a compromised host administrator and that someone has tested a restore.

Want help testing where a decoy alert goes? See Deception Check’s Design Partner program.

Related research · PHP2Shell probes: follow the destination beyond PHP