PHP2Shell (CVE-2026-109618, CVE-2026-109619): Patch Before the Bots Do
Two unauthenticated RCEs in PHP core, leaked a day before the patch, with exploit attempts already logged. Here is how to tell if you are exposed and what to do tonight.
Key takeaways
- PHP2Shell is two unauthenticated remote code execution flaws in PHP core, CVE-2026-109618 and CVE-2026-109619, both CVSS 9.2.
- Upgrade to PHP 8.5.12, 8.4.27, 8.3.36, or 8.2.35 and restart every worker.
- Apache mod_php is most exposed.
- Config mitigations reduce risk but PHP says only the update is a complete fix.
On this page
Upgrade PHP to 8.5.12, 8.4.27, 8.3.36, or 8.2.35 and restart every worker, today. PHP2Shell is two unauthenticated remote code execution bugs in PHP core, rated CVSS 9.2. Apache mod_php is the worst case: the confirmed exploit fires during request startup, before your application code runs, so framework hardening never gets a say.
Verified against
PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35 (released October 21, 2026), against the php-src GitHub security advisories, the php.net release notes and CISA's Known Exploited Vulnerabilities catalog. Checked October 2026.
The timing is what makes this one dangerous. The CVE IDs leaked on October 20, PHP moved its release up a day to October 21, and Amazon's honeypots logged an exploitation attempt on October 17, four days before disclosure. Palo Alto's Unit 42 counts 22.6 million potentially exposed instances. CISA added both CVEs to its Known Exploited Vulnerabilities catalog on October 26. Everything below is updated as of October 31, 2026.
Check if you're exposed in 30 seconds
You need two facts per host: the PHP version and the SAPI (how PHP is wired to the web server).
# Version and SAPI of the PHP your web server actually runs
$ php -v | head -1
$ php -r 'echo PHP_VERSION, " ", PHP_SAPI, PHP_EOL;' # CLI view; check the web SAPI below
$ apachectl -M 2>/dev/null | grep -i php # php_module => mod_php
$ systemctl list-units --type=service | grep -i fpm # php8.x-fpm => PHP-FPM
# Distro packages backport fixes without changing the upstream version
$ apt-cache policy php8.2-fpm libapache2-mod-php8.2 # Debian/Ubuntu
$ rpm -q --changelog php-fpm | grep -m3 -i 'CVE-2026-1096' # RHEL family
# Containers: check the image, not the host
$ docker run --rm --entrypoint php your-app:prod -v
Two traps. First, the CLI php binary can differ from the one Apache or FPM loads, so confirm the web SAPI with a phpinfo() page on a private route or the FPM binary (php-fpm8.3 -v). Second, Debian, Ubuntu, and RHEL patch inside their own package versions. php -v printing 8.2.28 can be safe on Debian if the package changelog lists the CVEs. Trust the vendor advisory over the upstream number.
What the two bugs actually do
CVE-2026-109618, "RCE via OOM corruption." PHP handles out-of-memory errors with longjmp(), jumping straight out of whatever it was doing. In some spots, request memory is reachable before it is fully initialized. If any allocation fails right then, request shutdown later reads that half-built state as valid. The advisory lists several ways a multipart request can hit OOM while staying under post_max_size, including bodies with no usable content length and parts with unbounded header counts.
CVE-2026-109619, "RCE via stale global state." A worker keeps global pointers (error, session, extension state) into memory owned by one request. If a fatal error interrupts the cleanup, the request's heap is freed while the global still points at it. The next request on the same worker can touch that stale pointer before initialization resets it: a use-after-free. Session upload progress, which is on by default and needs no session_start(), is one confirmed path, and it was validated against both mod_php and FPM.
Which setups are most exposed
| Setup | Status per PHP advisory | Why |
|---|---|---|
| Apache mod_php | Confirmed RCE, default limits | Shutdown runs even when startup fails; exploit runs pre-application |
| FrankenPHP | Likely susceptible | Same unconditional shutdown as mod_php |
| PHP-FPM | Confirmed via session upload path | Child exits on failed startup, but app-time paths remain |
| LiteSpeed (LSAPI) | Conditional, unvalidated | Models FPM; app-specific exploits likely |
| php-cgi | At risk | PHP lists every web-serving install as affected |
If you can't patch tonight
PHP is blunt that updating is the only complete fix. If a rollout needs a day, apply these administratively (so applications can't override them) and restart every worker:
; /etc/php/8.3/fpm/pool.d/www.conf (temporary, remove after patching)
php_admin_value[memory_limit] = -1
php_admin_flag[session.upload_progress.enabled] = off
; PHP 8.5 only, removes the confirmed CVE-2026-109619 producer:
; php_admin_flag[fatal_error_backtraces] = off
For mod_php, use php_admin_value and php_admin_flag in the Apache vhost instead. Then:
- Cap request size at the front. Set
client_max_body_sizein nginx orLimitRequestBodyin Apache, and block multipart bodies on routes that never take uploads. - Watch memory at the container level.
memory_limit = -1removes the predictable OOM the exploit relies on, but it also removes PHP's own guard. Give each worker its own cgroup limit if you can. - Do not use
RLIMIT_ASorRLIMIT_DATA. It sounds like the safe way to cap memory. It makesmmap()fail, which triggers exactly the vulnerablelongjmp()path. - Optionally set
pm.max_requests = 1if you can afford the performance hit, so no worker serves a second request with stale state.
Cloudflare and Palo Alto shipped WAF signatures. Cloudflare calls its rules "a STOPGAP only," and they cover only traffic that passes through them. Treat a WAF as time bought, not a fix.
End-of-life PHP has no release
PHP 7.4 through 8.1 get source backports only, with no official binaries. PHP 5.4 through 7.3 get nothing, and the vulnerable code goes back to 5.4.0. CERT-EU recommends isolating internet-facing applications from public networks if you can't update them, and on 8.1 or earlier "can't update" is your default. If you inherited a PHP 7.4 box, finish that upgrade this week. We wrote up how to triage vulnerabilities in production when the fix is not one command.
What attackers are doing
This is not theoretical. Amazon reports broad exploitation attempts, and Cloudflare and watchTowr both report attacks against unpatched hosts. On October 24 a researcher reported three honeypots running PHP 8.4.26 successfully exploited via CVE-2026-109618, and on October 26 Trail of Bits said working exploits hit pre-patch PHP 8.2 through 8.5. Dream Research has reported a campaign chaining PHP2Shell with a Linux kernel privilege escalation, which Unit 42 says it verified. Both CVEs have been in CISA's KEV catalog since October 26, with a federal remediation deadline of October 29. If you are reading this unpatched, assume you are already being scanned.
After patching, check for signs of earlier compromise: unexpected child processes of php-fpm or httpd (shells, curl, wget), new files in web roots, and outbound connections from web workers. Rebuild and rescan container images too, since a patched host running an old php:8.3-apache image is still exposed. Our notes on container vulnerability scanning cover catching stale base images in CI.
Questions people ask
Is PHP-FPM affected by PHP2Shell?
Yes. FPM is harder to exploit than mod_php because it kills a child whose request startup fails, but PHP validated the session upload progress path for CVE-2026-109619 against a normal dynamic FPM pool. Upgrade FPM installs to a fixed release, and until then disable session.upload_progress.enabled administratively and restart the pool.
Does a WAF protect me from CVE-2026-109618?
Partly, and only temporarily. Cloudflare and Palo Alto deployed signatures for known vectors, but Cloudflare describes its rules as a stopgap, and they only see traffic routed through them. The bug lives in how PHP parses requests, so new variants can slip past signatures. Patch PHP and use the WAF to buy hours, not weeks.
Is PHP 7.4 vulnerable to PHP2Shell?
Yes. The affected code path dates back to PHP 5.4.0. PHP 7.4 through 8.1 receive source backports only, with no official release, and 7.3 and earlier receive nothing. Check whether your distro or an LTS vendor shipped a patched package; otherwise upgrade to a supported branch or take the host off the internet.
The decision, concretely
- Running mod_php on the public internet? Patch now, and if you can't, take the site offline until you can.
- PHP-FPM behind nginx? Patch this week at the latest; apply the pool mitigations today.
- Distro packages? Patch through the distro and verify the CVE in the changelog, not the version string.
- PHP 8.1 or older? No upstream fix; get on a supported branch or a vendor's LTS package.
The call we'd make
Treat PHP2Shell like a weekend-cancelling event if you run mod_php or FrankenPHP on the internet, and as a this-week patch for FPM. Upgrade, restart every worker, rebuild images, and then look back at your logs since October 17. The config mitigations are real but partial; anyone who stops there is betting their uptime on a bug class that has already produced two CVEs this month.
Sources
Cite this article
Kiril Urbonas (2026). PHP2Shell (CVE-2026-109618, CVE-2026-109619): Patch Before the Bots Do. DevOpsNess. Updated 2026-11-06. https://www.devopsness.com/blog/php2shell-cve-2026-109618-109619
[PHP2Shell (CVE-2026-109618, CVE-2026-109619): Patch Before the Bots Do](https://www.devopsness.com/blog/php2shell-cve-2026-109618-109619) (Kiril Urbonas, DevOpsNess, updated 2026-11-06)
More from DevOps
Explore more articles in this category
kaniko CVE-2026-105769: Lookalike ACR Hosts Get Your Azure Token. Rebuild on v1.25.20
A FROM line pointing at a host like x.azurecr.io.attacker.com is enough to make kaniko hand the build's Entra ID token to that host. Which kaniko forks and images are affected, how to find Azure identities in your build pods, and what to rotate.
Sign Container Images With Cosign: Keyless GitHub Actions Workflow, SBOM and In-Cluster Checks
One workflow that builds, signs keylessly with GitHub OIDC, attaches an SBOM and SLSA provenance, plus the Kyverno policy that refuses anything unsigned. Copy it, change three names, ship it.
GitLab CVE-2026-110264: A Second CVSS 10 RCE in Nine Days. Patch to 19.5.1
An unauthenticated regex-compiler RCE reaching back to GitLab 7.0, no workaround, and backports on older branches that fix only the critical bug. The exact version per branch and the commands to get there.
You might have missed
Evergreen posts worth revisiting.