Skip to navigation

Threat reportVulnerabilityTL-2026-4686

PHP2Shell (CVE-2026-109618/CVE-2026-109619) + Fragnesia (CVE-2026-46300) Chain Roots 25 Servers, Breaches U.S. Government-Affiliated Network

criticalACTIVE

A suspected China-nexus operator chained the PHP2Shell unauthenticated RCE flaws with the Fragnesia Linux kernel LPE against internet-exposed firewall management interfaces, rooting 25 servers in the U.S., Japan, Taiwan and South Korea. The actor reached the internal network of a U.S. government-affiliated organization via root PHP web shells and an 8-port reverse SSH tunnel, then began Active Directory reconnaissance and password spraying.

Key facts

IdentifiersTL-2026-4686, CVE-2026-109618, CVE-2026-109619, CVE-2026-46300, CVE-2026-115560
Also known asPHP2Shell, Fragnesia
SeverityCRITICAL
CVSS9.2
First disclosed
Exploitation statusLinked CVE exploited in the wild; Listed in CISA KEV: CVE-2026-109618 (added ), CVE-2026-109619 (added )
EPSSCVE-2026-109618: 99.9% (99.9th percentile); CVE-2026-109619: 99.9% (99.9th percentile); CVE-2026-46300: 2.4% (83.7th percentile)
Affected productsPHP Group PHP (session.upload_progress; mod_php, PHP-FPM, php-cgi, FrankenPHP, LiteSpeed); Linux Linux kernel XFRM ESP-in-TCP (Fragnesia); Undisclosed large firewall vendor Firewall appliance appweb management server (PHP 8.0.30)
Weaknesses (CWE)CWE-416, CWE-400, CWE-755, CWE-824, CWE-1325
Nation-state nexusChina
MalwarePHP, iox
ToolsClaude Code, FOFA, aiscan, iox
MITRE ATT&CK techniquesT1021.001 Remote Services, T1037.004 RC Scripts, T1059.004 Unix Shell, T1068 Exploitation for Privilege Escalation, T1069.002 Domain Groups, T1087.002 Domain Account, T1090 Proxy, T1090.001 Proxy, T1110.003 Password Spraying, T1136.002 Domain Account, T1190 Exploit Public-Facing Application, T1505.003 Web Shell and 10 more in the attack flow
MotivationESPIONAGE
Target sectorsgovernment administration, network-security-appliances
Target regionsNorth America, East Asia, japan, taiwan, south korea
Detection rules9 (1 shown free)
Sources10 cited sources

Detection coverage

As of 2026-11-01, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-4686 across Splunk SPL, Microsoft KQL and Sigma, covering 24 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.

Free detection: Fragnesia Aftermath: Passwordless su to Root by Web Service Account (/etc/pam.d/su pam_permit.so)

TL-2026-4686: after PHP2Shell yields command execution as nobody, the actor used Fragnesia (CVE-2026-46300) to overwrite the page-cache copy of /etc/pam.d/su with a pam_permit.so configuration and ran su root without a password. The page-cache write does not change the file on disk, so file-integrity monitoring misses it; the su syslog record (to root) from a web or service account with no tty is the reliable behavioural signal.

  • Languages: Splunk SPL, Microsoft KQL, Sigma
  • ATT&CK: T1068, T1556.003
  • Data sources: linux_secure (index=linux, sourcetype=linux_secure), syslog (auth facility), Syslog table (Microsoft Sentinel), linux auth logsource
  • Severity: critical
  • Confidence: high

Splunk SPL

index=linux (sourcetype=linux_secure OR sourcetype=syslog) "to root"
| rex field=_raw "su[^:]*: .to (?<target_user>[A-Za-z0-9_.-]+). (?<source_user>[A-Za-z0-9_.-]+) on (?<tty>[^ ]+)"
| where target_user="root" AND match(source_user,"^(nobody|www-data|apache|www|http|daemon|nginx)$")
| stats count min(_time) AS first_seen max(_time) AS last_seen values(tty) AS ttys BY host source_user target_user
| convert ctime(first_seen) ctime(last_seen)

Microsoft KQL

Syslog
| where TimeGenerated > ago(24h)
| where ProcessName =~ 'su' and SyslogMessage has 'to root'
| extend TargetUser = extract(@'to ([A-Za-z0-9_.-]+). ([A-Za-z0-9_.-]+) on ([^ ]+)', 1, SyslogMessage), SourceUser = extract(@'to ([A-Za-z0-9_.-]+). ([A-Za-z0-9_.-]+) on ([^ ]+)', 2, SyslogMessage), Tty = extract(@'to ([A-Za-z0-9_.-]+). ([A-Za-z0-9_.-]+) on ([^ ]+)', 3, SyslogMessage)
| where TargetUser == 'root' and SourceUser in~ ('nobody', 'www-data', 'apache', 'www', 'http', 'daemon', 'nginx')
| summarize Events = count(), FirstSeen = min(TimeGenerated), LastSeen = max(TimeGenerated), Ttys = make_set(Tty, 5) by Computer, SourceUser, TargetUser

Sigma

title: Web Service Account Switches to Root via su (Fragnesia pam_permit.so Bypass)
id: 8a3e91b4-2c57-4d08-b6f1-0e9d7a5c3b12
status: experimental
description: Detects su to root by a web or daemon service account. In the PHP2Shell and Fragnesia (CVE-2026-46300) chain the actor replaced the page-cache copy of /etc/pam.d/su with a pam_permit.so configuration, so su root succeeded from the nobody web shell without a password.
references:
  - https://dreamgroup.com/blog/chained-php2shell-exploit-roots-25-servers
  - https://tenable.com/blog/fragnesia-cve-2026-46300-faq-about-new-linux-kernel-xfrm-esp-in-tcp-priv-esc
author: Threadlinqs Team
date: 2026-10-28
tags:
  - attack.privilege-escalation
  - attack.t1068
  - attack.credential-access
  - attack.t1556.003
  - cve.2026-46300
logsource:
  product: linux
  service: auth
detection:
  keywords:
    - '(to root) nobody on'
    - '(to root) www-data on'
    - '(to root) apache on'
    - '(to root) www on'
    - '(to root) http on'
    - '(to root) daemon on'
    - '(to root) nginx on'
  condition: keywords
falsepositives:
  - Misconfigured service wrappers that legitimately su from a service account (rare; verify against change records)
level: critical

Known false positives:

  • Service-account maintenance scripts that su to root by design
  • Legacy cron or init wrappers running su from daemon accounts

Tuning notes: Extend the service-account list to the accounts your web tier runs as (appweb or vendor-specific users). A hit with tty 'none' is a strong web-shell indicator. Also run an out-of-band check that /etc/pam.d/su contains no pam_permit.so, since the Fragnesia page-cache write is invisible to on-disk integrity tools (drop caches or compare against the package hash).

8 more detection rules (Splunk SPL, Microsoft KQL, Sigma) · unlock with TL Creds. Sign up free to unlock the detection rules or query them with the Threadlinqs MCP server (Purple)
  • PHP/Web Server Process Spawning Shell, Recon or Credential-Harvest Commands (PHP2Shell RCE as nobody) · unlock with TL Creds
  • PHP Web Shell Dropped on Web Server (103-byte Root Web Shell, Web-Process or Root-Shell Written) · unlock with TL Creds
  • Password Spray and AD User Enumeration from a Single Source (Firewall Management Server Post-Compromise) · unlock with TL Creds
  • PHP/Web Server Crash in [indicator] (PHP2Shell Custom Exploit Memory Corruption Attempt) · unlock with TL Creds
  • Systemd Service or rc.local Boot Hook Written Outside Package Management (Tunnel Persistence) · unlock with TL Creds
  • Rogue Domain Account Creation and AD Group Enumeration (MachineAccountQuota Abuse, Backdoor Users in Every OU) · unlock with TL Creds
  • PHP2Shell Oversized Multipart POST / Scanner Fan-out to PHP Endpoints (CVE-2026-109619 / CVE-2026-109618) · unlock with TL Creds
  • Reverse SSH Tunnel, iox SOCKS Proxy and AI Attack Tooling (aiscan, DeepSeek/Kimi API) from Firewall or Management Host · unlock with TL Creds

Indicators of compromise

As of 2026-11-01, Threadlinqs Intelligence tracks 24 indicators of compromise for TL-2026-4686.

24 indicators of compromise · unlock with TL Creds. Compare plans

Technical analysis

Dream researchers recovered artifacts from attacker infrastructure showing a custom, non-public weaponization of PHP2Shell (CVE-2026-109618 and CVE-2026-109619) paired with Fragnesia (CVE-2026-46300) against the appweb management server of an undisclosed large firewall vendor. Targets ran PHP 8.0.30 (7.4.33 also appears in the artifacts) with SELinux httpd_t confinement. Dream does not attribute the activity to a named actor, but cites multiple China-nexus indicators: targeting limited to the U.S., Japan, Taiwan and South Korea, infrastructure on UTC+8, use of the FOFA search engine, the Chinese-developed iox tunneling tool, DeepSeek and Kimi LLM configurations, and Chinese-language prompts in aiscan logs.

PHP2Shell is a pair of memory-safety flaws disclosed and patched by the PHP project on 2026-10-21 (fixed in PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35, with backports for 7.4, 8.0 and 8.1). CVE-2026-109619 is a use-after-free (CWE-416) in the session.upload_progress feature, where cached zval pointers dangle after session hash tables are replaced. CVE-2026-109618 is memory corruption after an out-of-memory condition forces a longjmp that skips cleanup (CWE-400). Both are rated CVSS v4.0 9.2. Per Dream, the exploit ran in stages: memory exhaustion of about 383MB via a multipart POST with crafted filenames, an out-of-bounds read leaking the libphp.so base address through a Set-Cookie header, a stale-reference setup of about 500MB, and a forged HashTable whose pDestructor points to zif_system. This yielded command execution as the nobody user.

Privilege escalation used Fragnesia (CVE-2026-46300, CVSS 7.8), a Linux kernel XFRM ESP-in-TCP flaw disclosed on 2026-05-13 by William Bowling (V12 Security). skb_try_coalesce() fails to propagate SKBFL_SHARED_FRAG, allowing writes into the page cache of readable files. The actor used a per-byte page-cache write to replace /etc/pam.d/su (774 bytes) with a pam_permit.so configuration, then ran su root without a password. A 103-byte root PHP web shell was planted and verified with the id command. Persistence was a systemd service plus an rc.local boot hook, and an 8-port reverse SSH tunnel with a SOCKS5 proxy (iox is listed among the tools) gave access to the internal network.

The campaign funnel was about 19,000 firewalls found through FOFA, about 5,000 alive per a custom scanner (200 threads), 240 filtered targets (target countries, PHP 8.0.30), and 24 firewalls exploited. Of these, 22 were parked with web shells and 2 government targets were prioritized. One internal firewall management server was also rooted, giving 25 compromised servers in total. Eleven JSON version profiles held SHA256 hashes, ELF build IDs and exact memory offsets for each tested firewall OS version. About 40 custom post-exploitation scripts showed AI-assisted development, and Claude Code and aiscan (with DeepSeek and Kimi) were found on the infrastructure.

Post-compromise, from the management server the actor decrypted the LDAP service account password from the encrypted device config, enumerated users and groups, and sprayed passwords against all users (the domain had lockoutThreshold=0). The actor also attempted backdoor user creation in every OU and machine account creation via the default MachineAccountQuota, analyzed AD ACL chains, and probed Keycloak for SAML signing key extraction and authentication bypass. Outcomes of the spray, backdoor and key-extraction attempts were not recovered. Harvested material included HA replication keys, TLS certificate keys and device authentication keys. CISA added CVE-2026-109618 and CVE-2026-109619 to the KEV catalog on 2026-10-28 with a federal due date of 2026-10-29. No network IOCs (IPs, domains, hashes) were published in the sources reviewed, so none are listed.

MITRE ATT&CK techniques by tactic:

Lateral Movement

T1021.001 Remote Services

Persistence

T1037.004 RC Scripts; T1136.002 Domain Account; T1505.003 Web Shell; T1543.002 Systemd Service

Execution

T1059.004 Unix Shell

Privilege Escalation

T1068 Exploitation for Privilege Escalation

Discovery

T1069.002 Domain Groups; T1087.002 Domain Account

Command and Control

T1090 Proxy; T1090.001 Proxy; T1572 Protocol Tunneling

Credential Access

T1110.003 Password Spraying; T1552 Unsecured Credentials; T1552.001 Unsecured Credentials; T1552.004 Unsecured Credentials; T1556.003 Pluggable Authentication Modules

Initial Access

T1190 Exploit Public-Facing Application

Resource Development

T1587.004 Exploits; T1588.007 Artificial Intelligence

Reconnaissance

T1595.002 Vulnerability Scanning; T1596.005 Scan Databases

Remediation

Patches

  • Upgrade PHP to 8.5.12, 8.4.27, 8.3.36 or 8.2.35, or apply vendor backports for 7.4, 8.0 and 8.1
  • Apply Linux kernel updates containing the May 13, 2026 Fragnesia fix (CVE-2026-46300)
  • Apply the firewall vendor's fixed release once available

Immediate actions

  • Remove internet exposure of firewall and appliance management interfaces; restrict them to dedicated management networks
  • Hunt for 103-byte PHP web shells in web-accessible directories of firewall management servers
  • Hunt for unexpected reverse SSH tunnels, SOCKS5 proxies and iox-style tunnel processes originating from firewall and management hosts
  • Check /etc/pam.d/su for replacement with pam_permit.so and for unexpected systemd units or rc.local entries
  • Rotate LDAP service account, HA replication, TLS and device authentication credentials stored on potentially compromised appliances

Workarounds

  • Set session.upload_progress.enabled=0 (partial; does not cover all variants)
  • Set memory_limit=-1 and reduce post_max_size for CVE-2026-109618 (partial)
  • Blacklist esp4, esp6 and rxrpc modules for Fragnesia: rmmod esp4 esp6 rxrpc and install them as /bin/false in modprobe.d

Longer-term hardening

  • Set a non-zero AD lockoutThreshold and reduce ms-DS-MachineAccountQuota
  • Audit AD ACLs and review for unexpected user or machine accounts in every OU
  • Validate SAML IdP certificate checking and review Keycloak for signing key access
  • Monitor for anomalous outbound SSH from perimeter devices

Affected and fixed versions

  • PHP Group — PHP (session.upload_progress; mod_php, PHP-FPM, php-cgi, FrankenPHP, LiteSpeed)
    Vulnerable versions: All PHP installations serving web requests before the October 21, 2026 releases; 8.0.30; 7.4.33
    Fixed in: 8.5.12; 8.4.27; 8.3.36; 8.2.35; Backports for 7.4, 8.0, 8.1
  • Linux — Linux kernel XFRM ESP-in-TCP (Fragnesia)
    Vulnerable versions: Kernels without the May 13, 2026 netdev fix (Ubuntu, RHEL, openSUSE, CentOS Stream, Debian, Gentoo reported affected)
    Fixed in: Kernels with the May 13, 2026 fix, e.g. AlmaLinux 5.14.0-611.54.4.el9_7
  • Undisclosed large firewall vendor — Firewall appliance appweb management server (PHP 8.0.30)
    Vulnerable versions: Versions bundling vulnerable PHP; vendor not named by researchers

Timeline

Last updated · 3 updates · revalidated 3× · latest source

  • Fragnesia (CVE-2026-46300) Linux kernel XFRM ESP-in-TCP LPE disclosed by William Bowling (V12 Security); fix posted to netdev and public PoC released.
  • Wiz discovers the PHP memory-safety flaws using frontier AI models; RCE confirmed against Drupal and WordPress on 2026-08-20/21.
  • Wiz reports the issues to the PHP security team; PHP acknowledges on 2026-09-21 and begins fix development.
  • Per Dream's analysis of recovered artifacts, the attacker's campaign infrastructure setup began.
  • First exploitation attempts recorded, four days before patches (Dream AWS honeypot observation).
  • Palo Alto Networks releases Frontier Virtual Patching coverage (Threat ID 543004) for Advanced Threat Prevention Plus customers.
  • PHP2Shell (CVE-2026-109618 and CVE-2026-109619) announced; discoverers credited as Calif, OpenAI and Wiz.
  • PHP project released 8.5.12, 8.4.27, 8.3.36 and 8.2.35 (plus backports for 7.4, 8.0, 8.1); Cloudflare reported early attacks and deployed WAF protections.
  • Palo Alto Networks confirms PAN-OS and Panorama management interfaces are affected (CVSS-B 8.7) and says it is not aware of malicious exploitation against its products; fixed builds staggered through 2026-10-30.
  • First U.S. government-affiliated firewall rooted at 18:14 UTC; it crashed after exploitation and was probed 240 times over the next two hours.
  • Second government firewall exploited at 10:09 UTC; at 11:35 UTC the attacker rooted the internal firewall management server through the tunnel and began Active Directory activity. Honeypots also saw 35 shell commands from two IPs against PHP 8.4.26 under mod_php.
  • Unit 42 retracts its 2026-10-23 post-exploitation claims against a European government agency, stating it cannot confirm initial access was tied to PHP2Shell.
  • Dream published its findings: 25 servers rooted (24 perimeter firewalls plus one internal management server) in the U.S., Japan, Taiwan and South Korea, with 22 firewalls still holding active web shells.
  • CISA added CVE-2026-109618 and CVE-2026-109619 to the KEV catalog with a federal remediation deadline of 2026-10-29.
  • The Register reports Unit 42 independently verified Dream's tooling and findings, and that the reverse tunnel into the U.S. government organization is likely still running.
  • The Hacker News, citing Dream, reports exploitation against internet-exposed PAN-OS management interfaces, including three U.S. government-affiliated systems (secondary attribution of the withheld vendor).

Update history

Sources cited

Other reports on CVE-2026-109618

Other reports on CVE-2026-109619

Previous and next in vulnerability

Further reading

Drafted by the Threadlinqs research pipeline from 10 sources; published ; last updated .

Threadlinqs Intelligence — Real-Time Threat Detection Platform

// threat_feed
$ sort --newest
Showing all threats

Live intelligence console

Threat level

Latest Threats