Defense tooling for the PHP "php2shell" security release of 2026-10-21:
- CVE-2026-109618 — RCE via OOM corruption (GHSA-w8hw-w9fx-j79w)
- CVE-2026-109619 — RCE via stale global state (GHSA-6wmf-4rv5-p3rp)
Affected: PHP 8.2 < 8.2.35 · 8.3 < 8.3.36 · 8.4 < 8.4.27 · 8.5 < 8.5.12. Updating is the only complete fix; these tools help you verify and detect.
EOL branches (7.4, 8.0, 8.1): the PHP security team published source
backports — these branches are not automatically "always vulnerable".
Verify your distribution's backport status; memory_hard_limit_percent
absence does NOT prove vulnerability (the 8.1 backport fixes both CVEs
without adding that directive).
| Component | Files | Description |
|---|---|---|
| Safechecks | 4 Python scripts | Behavioural vulnerability and configuration checks (local + remote + bulk scan). See safechecks/README.md for full documentation. |
| Suricata | 1 rules file (21 signatures) | HTTP request/response signatures for Snort/Suricata. Validated on real exploit traffic (mod_php + FPM). |
| ModSecurity | 1 config file (21 rules) | WAF rules on parsed variables. Includes response-side behavioural detection and SAPI-specific crash-signal documentation. |
| Nuclei | 4 templates (behavioural + surface + EOL + startup-surface) | |
| Sigma | 4 SIEM rules (HTTP exploit, PHP OOM, session decode, process creation) | Non-destructive behavioural detection with multi-path payloads. |
| Host hunt | 7 YARA rules, dated IOC manifest, Python runner and 25 regression tests | Read-only filesystem triage of webroots, PHP session files, temporary artifacts and decoded request captures. See host-hunt/README.md / Русский. |
# Single target:
python3 safechecks/cyberok_safecheck_cve_2026_109618.py -u https://host/
# Bulk scan (nuclei-style targets file + threads + JSON):
python3 safechecks/cyberok_safecheck_cve_2026_109618.py -l targets.txt -t 20 -o results.json
# Custom paths:
python3 safechecks/cyberok_safecheck_cve_2026_109618.py -u https://host/ --path /index.php --path /api/status
# CVE-2026-109619 — same interface:
python3 safechecks/cyberok_safecheck_cve_2026_109619.py -u https://host/
python3 safechecks/cyberok_safecheck_cve_2026_109619.py -l targets.txt -t 20 -o results.jsonpython3 safechecks/cyberok_safecheck_cve_2026_109618.py --php-cgi /usr/bin/php-cgi
python3 safechecks/cyberok_safecheck_cve_2026_109619.py --php-cli /usr/bin/phppython3 safechecks/cyberok_safecheck_postpatch_surface.py --php-cli /usr/bin/php# Suricata:
include: suricata/cyberok_php_up_2026.rules
# offline replay:
suricata -c suricata.yaml -S suricata/cyberok_php_up_2026.rules -k none -r capture.pcap -l out/
# ModSecurity:
Include modsecurity/cyberok_php_up_2026.conf
# Nuclei:
nuclei -t nuclei/cyberok_php2shell_detect.yaml -u https://target/| SID | Detection | Vector |
|---|---|---|
| 6101 | Chunked multipart POST | 109618-A (post_max_size bypass) |
| 6102 | Part-header amplification (100+) | 109618-B |
| 6103 | Generic header-run amplification | 109618-B (variant) |
| 6104 | Deeply nested input names (40+) | 109618 aux |
| 6105 | Empty-filename file control | 109618-C |
| 6106 | Serialized payload in UPLOAD_PROGRESS | 109619 (poisoning) |
| 6107 | Payload in multipart filename | 109619 (carrier) |
| 6108 | PHPSESSID as form field | 109619 (sid injection) |
| 6109 | 10+ consecutive Content-Disposition headers | Header-run anomaly; not a total multipart-part counter |
| 6110 | Handler-mismatch seed (pipe+serialized) | 109619 (handler mismatch) |
| 6111 | Set-Cookie PHPSESSID leak oracle | 109618 (response-side) |
| 6112 | Giant multipart filename (>4000 chars) | 109618 arm |
| 6113 | UPLOAD_PROGRESS field present (anomaly) | Generic |
| 6114 | Invalid PHPSESSID cookie (charset) | Generic |
| 6115 | Session-file path in URI (LFI trigger) | 109619 (include/LFI sink) |
| 6116 | OOM in response body | 109618 (response-side, vulnerable only) |
| 6117 | Session destruction failed in response | 109619 (response-side, vulnerable only) |
| 6118 | Failed to decode session object | 109619 (response-side) |
| 6119 | Post-exploit webshell access (p2s_*) | Post-exploit IoC |
| 6120 | Post-exploit marker (SHELL_ALIVE) | Post-exploit IoC |
| 6121 | Serialized integer-key collision sequence | Captured exploit request pattern (Corelight-derived) |
Request-side (phase 1–2): matching Suricata coverage on parsed variables + robust occurrence counting. Response-side (phase 3/4): behavioural detection via three-warning signature, OOM indicator, session-decode failure, and post-exploit markers. Includes SAPI-specific crash-signal taxonomy (mod_php vs FPM) and rejected-statuses troubleshooting guide.
Use filesystem evidence alongside the existing event-based Sigma rules. The host-hunt rules distinguish exact known malware sample hashes, public PoC content, suspicious session data and generic PHP execution patterns. A match requires investigation; it does not automatically prove execution or attribute initial access to these CVEs. Each IOC records its source and role.
From the repository root, using actual paths from the PHP instance or a mounted snapshot:
python3 -m venv host-hunt/.venv
host-hunt/.venv/bin/python -m pip install -r host-hunt/requirements.txt
host-hunt/.venv/bin/python host-hunt/hosthunt.py \
--webroot /mnt/server/var/www/html \
--session-dir /mnt/server/var/lib/php/sessions \
--artifact-dir /mnt/server/tmp \
> host-hunt/findings.jsonl
# Offline regression tests; no PHP or malware is executed:
host-hunt/.venv/bin/python -m unittest discover -s host-hunt -p 'test_*.py' -vThe runner outputs paths, hashes and rule metadata without session contents.
Exit 1 means findings (also inspect incomplete); 2 means incomplete without
findings; 0 means no match in the scanned scope. Unreadable, oversized,
changing, archived and symlinked files are reported as skipped. Filesystem
scanning does not cover process memory, deleted artifacts or Redis/Memcached
session stores. See the host-hunt guide for scope, limits
and log-search examples.
- Sigma telemetry: the HTTP rule needs
http_request_body; standard nginx combined access logs do not contain request bodies or cookies. Process rules require process-creation telemetry and backend field mapping. YARA supplies complementary disk evidence, not missing historical events. - PoC provenance:
p2s_8512_shell.phpandSHELL_ALIVEbelong to different public PoCs. Host-hunt detects them separately. Authorized checks and research fixtures may match; known test markers are reported, never automatically suppressed. A negative scan does not establish that a host was uncompromised. - Suricata counting caveat: occurrence counting (
{N,}) over variable-gap multipart units is unreliable under PCRE2 + inspect windows — the Suricata set uses unit signatures; robust counting lives in ModSecurity. - Post-patch relevance: rules for
PHP_SESSION_UPLOAD_PROGRESSpoisoning andPHPSESSID-as-form-field remain useful on patched versions (the form-SID ingestion path survives the 8.2.35 fix withcleanup=0). - Safe-checks are behavioural, not version checks: they conclude from the
signal itself;
php -vremains the authoritative version answer. - EOL branches (5.6–8.1): source backports exist (php.net) — verify your distro
the vulnerable behaviour on them (all confirmed vulnerable by direct
testing). Remote detection on ≤7.4 requires
display_startup_errors=Onon the target; on hardened production (both display flags Off) no patch-correlated signal reaches the HTTP response at all — headers, body and cookie behaviour are byte-identical between unpatched and patched builds, so INCONCLUSIVE there is by design (seesafechecks/README.mdfor the oracle hunt, incl. the fragile timing differential).
| Test | Result |
|---|---|
| Safechecks local (8.2.34 vs 8.2.35 ASan) | Both decisive (A/B) |
| Safechecks remote (php:8.2.34-apache vs php:8.2.35-fpm) | Both decisive |
| Bulk scan (2 targets, JSON) | Correct: 1 vulnerable + 1 patched |
| Suricata vs mod_php traffic | 9/10 request signatures fired, control silent |
| Suricata vs FPM traffic | 9/10 fired, control silent |
| Suricata new rules on FPM (6115-6120) | 4/6 fired (LFI, webshell IoC, 8512-seed, anomaly); 2 response-side need vulnerable build |
| ModSecurity patterns | Desk-verified against exact trigger bytes |
| phpMyAdmin retest | Technique validated (Assetnote) |
| ikow rules comparison | 10/12 covered by our set; 2 gaps (threshold-recon, version-inventory) noted |
| Host-hunt YARA | 7 rules compile without warnings in yara-python 4.5.4 |
| Host-hunt regression tests | 25/25 offline tests passed; synthetic positives/negatives and runner coverage/error semantics. No real-malware execution or clean-application FP corpus. |
(c) 2026 АО «Сайбер ОК» / JSC CyberOK. Provided for informational and research
purposes only; see LICENSE for the full copyright and usage notice (RU + EN).