Skip to content

About

Mitigations for PHP installs that cannot update yet: CVE-2026-109618 / CVE-2026-109619 (php2shell) and CVE-2026-115560. Edge rules, auth before PHP, request limits, PHP settings. Unofficial. The real fix is updating PHP.

Topics

Resources

Stars

9 stars

Watchers

0 watching

Forks

Latest commit

 

History

14 Commits

Folders and files

Repository files navigation

php2shell-hardening

Unofficial edge rules that stop multipart/form-data requests before they reach PHP, as a stopgap for CVE-2026-109618 and CVE-2026-109619 ("php2shell") on installations that cannot be updated yet. Covers Nginx, Apache, ModSecurity, Caddy, lighttpd, LiteSpeed, FrankenPHP, Cloudflare and php.ini, with a test suite. The session.upload_progress setting here is also the mitigation PHP asks for against CVE-2026-115560, a third issue published on 23 October 2026 that has no release yet.

If you run PHP 7.4, 8.0 or 8.1, look first for a vendor package with the fix: Ubuntu Pro for 20.04, Red Hat and AlmaLinux for 8 and 9, cPanel and CloudLinux for hosting, or a build from the backport branches. Those packages keep the old version number, see "Getting a fixed package".

These rules are provisional. They block legitimate uploads and multipart forms, so use them only on routes that do not need uploads and try them in staging first. The PHP project says that only an update is a complete fix, and that none of the configuration measures below make an installation safe.

What PHP published

On 21 October 2026 PHP released 8.5.12, 8.4.27, 8.3.36 and 8.2.35 and published an advisory with two critical issues, CVSS v4.0 9.2:

  • CVE-2026-109618, RCE via OOM corruption: GHSA-w8hw-w9fx-j79w. An out-of-memory failure can interrupt PHP while request data is only partly initialized. A complete exploit was demonstrated on Apache mod_php; for FPM it is source analysis only.
  • CVE-2026-109619, RCE via stale global state: GHSA-6wmf-4rv5-p3rp. Request state can survive after its memory is freed and be used by a later request on the same worker. session.upload_progress is a confirmed path, on mod_php and on FPM.

PHP 8.1 and older no longer get updates from the project. For 7.4, 8.0 and 8.1 the security team published minimal source backports for people who build from source: 7.4, 8.0, 8.1. Linux distributions and LTS vendors received the patches in advance, so check your vendor first. Several of them have already shipped the fix for the old branches (Ubuntu Pro for 20.04, RHEL and AlmaLinux, cPanel, CloudLinux, see below). Packages with a backport keep the old version number.

23 October 2026: CVE-2026-115560

Two days after the advisory PHP published a note on a third RCE vector, CVE-2026-115560, again through session upload progress. There is no release for it: PHP calls the scenario unlikely, handles it in the normal bugfix cycle and will publish the full advisory with the next release. No affected-version list yet, and the patched versions of 21 October do not cover it.

It needs three conditions at the same time:

  • session.serialize_handler set to php_binary or php_serialize in the configuration,
  • the application switching it to php with ini_set() before session_start(),
  • session upload progress still enabled.

The mitigation is the one already in this repository, session.upload_progress.enabled = 0. If you applied it for CVE-2026-109619 on 21 October there is nothing else to do: keep it off until the next bugfix release. Applications that never change the serialize handler at runtime can leave the feature on, so check the code before deciding:

grep -rn --include='*.php' -E "ini_set\s*\(\s*['\"]session\.serialize_handler" /var/www

Getting a fixed package

A vendor backport is the upstream fix applied to the old code without a new PHP version. The package keeps its number, so php -v says nothing about whether the fix is there, and scanners that compare only versions report patched systems as vulnerable. Check the package changelog for the CVE instead; the commands are in each section.

Ubuntu 20.04

php7.4 is fixed in 7.4.3-4ubuntu2.29+esm5 (USN-8948-2), available only with Ubuntu Pro. On 20.04 php7.4 is in main, so the fix is in esm-infra. The same notice covers 18.04 (php7.2 7.2.24-0ubuntu0.18.04.17+esm14, Pro) and 16.04 and 14.04 (php7.0 ...+esm22, php5 ...+esm18, Legacy Support add-on). Releases that are still supported get the fix without a subscription, from USN-8948-1: 22.04 php8.1 8.1.2-1ubuntu2.28, 24.04 php8.3 8.3.6-0ubuntu0.24.04.13, 26.04 php8.5 8.5.4-0ubuntu1.5. To check 20.04 after pro attach and apt upgrade:

dpkg -s php7.4-cli | grep ^Version    # 7.4.3-4ubuntu2.29+esm5 or later

The free Ubuntu Pro subscription covers up to 5 physical machines, with unlimited VMs and containers on them (50 for official Ubuntu community members). Canonical words the eligibility in different ways. ubuntu.com/pro says personal use, the 2022 announcement and the FAQ say personal and small-scale commercial use, and the personal terms allow one account per company or legal entity, ask you to name the company when you register on its behalf and cap it at 5 physical systems. None of them says what happens with a machine that belongs to a client or an employer, so ask Canonical before relying on it for those. Going over the limit is not blocked, but Canonical counts the active machines, and sudo pro detach frees a slot.

On a machine that is not attached, the table printed by pro status lists the services that are available, not the ones that are active. Check attached in pro status --format json.

If PHP 7.4 came from the ondrej PPA, its packages carry epoch 1: (1:7.4.33-20+ubuntu20.04.1) and apt sees them as newer than Ubuntu's: it never moves to the ESM build on its own, so install the ESM versions explicitly. Extensions taken from the PPA (apcu, imagick and similar) come from other source packages with their own versions, so check each of them too. The PPA removed its focal packages in June 2025 and the downloads now return 404, so save what is installed before switching, or there is no way back:

apt install dpkg-repack
dpkg-query -W -f='${db:Status-Abbrev} ${binary:Package}\n' 'php7.4*' | awk '$1 == "ii" { print $2 }' | xargs dpkg-repack

ondrej/sury packages

On 21 October php7.4 1:7.4.33-31 was published for Debian 11, 12 and 13 and Ubuntu 22.04, 24.04 and 26.04. Its changelog lists "Avoid state corruption via bailout" for CVE-2026-109618 and CVE-2026-109619, backported from 8.2.35. There is no such build for Ubuntu 20.04 or Debian 10.

When you update from these repositories, upgrade only the packages built from the PHP source you run. Upgrading the PECL extension packages (php8.2-apcu, php8.2-imagick, php8.2-memcache) pulls in the php-* metapackages, which install PHP 8.5 and switch /usr/bin/php to it.

apt-get install --only-upgrade $(dpkg-query -W -f='${db:Status-Abbrev} ${binary:Package} ${source:Package}\n' | awk '$1 == "ii" && $3 == "php8.2" { print $2 }')
# if /usr/bin/php has already moved
update-alternatives --set php /usr/bin/php8.2

Debian

Debian 12 has php8.2 8.2.35-1~deb12u1 (DLA-4848-1) and Debian 13 php8.4 8.4.27-1~deb13u1 (DSA-6574-1), as listed by the Debian security tracker. The tracker has no data for php7.4 on Debian 11. A third-party summary reports a Freexian ELTS (paid extended support) advisory, ELA-1853-1, with php7.4 7.4.33-1+deb11u14; confirm it with Freexian before relying on it. The free alternative is the sury.org build above.

Red Hat, AlmaLinux and derivatives

RHEL 8 has the fix in the php:7.4 module, RHSA-2026:87080, php 7.4.33-8, published on 21 October together with ten other CVEs. AlmaLinux announced the same day, on its blog, patched packages for every PHP stream on 8, 9 and 10. For RHEL 7 the bulletin RHSB-2026-013 lists session.upload_progress.enabled = 0 as the only change required. Check a package with:

rpm -q --changelog php-fpm | grep -E 'CVE-2026-10961[89]'

Hosting panels

CloudLinux, according to its blog: ea-php 5.2 to 8.5 fixed on 21 October at 20:55 UTC; alt-php 5.2 to 5.6, 7.0 to 7.4 and 8.0 to 8.5 in the stable repositories of CloudLinux 7, 8 and 9 by 00:07 UTC on 22 October; on CloudLinux 10 only 5.2 is in the normal repository and the others come from the ELS repository, which has to stay enabled; CloudLinux 6 is not confirmed. Its security metadata was not published at that point, so yum --security does not list these packages: use a plain dnf upgrade 'alt-php*' or 'ea-php*', then restart PHP.

rpm -q --changelog alt-php74-common | grep -E 'CVE-2026-(109618|109619)'
rpm -q --changelog ea-php74-php-cli | grep -E 'CVE-2026-(109618|109619)'

cPanel EasyApache 4 is reported to carry backports for ea-php74, ea-php80 and ea-php81 (7.4.33-19, 8.0.30-12, 8.1.34-5, EA-13613, EA-13612, EA-13611) by ComputingForGeeks, not by a cPanel note. ea-php73 and older have no fix; TuxCare ELS is a paid option, and its CVE tracker says whether php2shell is covered. The Plesk and DirectAdmin forum threads of 21 October gave no date and did not name 7.4, 8.0 or 8.1, so ask the vendor or the host which package contains the fix and for the changelog line with CVE-2026-109618.

Building from a backport branch

Without a vendor package, build from the branch of the PHP project. The 7.4 branch differs from the 7.4 code by one security commit dated 18 October 2026 (17 files, 240 lines added, 59 removed), and the 8.0 and 8.1 branches work the same way. Reuse the options of the PHP you run (php -i | grep 'Configure Command') and install into a separate prefix:

git clone --depth 1 --branch PHP-7.4-CVE-2026-109618-109619 https://github.com/php/php-src.git
cd php-src
git log -1 --format='%h %ad %s'
./buildconf --force
./configure --prefix=/opt/php-7.4-patched <options of the current PHP>
make -j"$(nproc)"
sudo make install

The branch has the two fixes and nothing else, and 7.4 has been out of support since November 2022, so other known flaws stay open and every later update is yours. Rebuild the PECL extensions with the phpize of the new prefix, point the FPM unit at the new binary, try it in staging with error_reporting = E_ALL, and keep the commit hash and the build log: php -v does not prove the patch.

Docker

Images built FROM php:8.1-fpm or older tags will not get the fix: there is no upstream release for those branches. Move to an 8.2.35 or later tag, or build from a backport branch.

If you cannot update

The risks, the cases where the flaws trigger and why a login or a missing upload form does not protect you are covered in PHP2Shell, what to do if you can't update PHP. The steps below, in order, from the strongest to the weakest.

  1. Use a vendor package that contains the fix (see "Getting a fixed package"), or build from one of the backport branches above.
  2. Try the smallest jump to a supported branch in staging, with error_reporting = E_ALL.
  3. Put authentication in front of PHP, so an unauthorized request is rejected before the engine reads its body. Application-level login does not help: PHP parses the request before your script runs.
  4. Limit request size at the edge and block multipart where it is not needed (the rules in this repository).
  5. Apply the PHP settings below, with a memory cap outside PHP.

Then check the server with test/audit-server.sh (see "Audit a server" below).

Authentication before PHP

Ready-to-edit files are in nginx/auth-before-php.conf.example and apache/auth-before-php.conf.example. In short, Nginx with basic auth or an allow-list:

location / {
    auth_basic           "restricted";
    auth_basic_user_file /etc/nginx/.htpasswd;
    # or: allow 203.0.113.0/24; deny all;
}

Apache 2.4:

<Location "/">
    Require ip 203.0.113.0/24
</Location>

Request limits at the edge

PHP asks for limits on decoded request size, field count, multipart part count, header count and nesting depth, enforced also on requests without a usable Content-Length. Nginx and Apache cover only part of that:

client_max_body_size 8m;
LimitRequestBody 8388608
LimitRequestFields 50

With Nginx in front of PHP-FPM the body is buffered and forwarded with a Content-Length; check that fastcgi_request_buffering is not off. Per-part header and part counts are not configurable in either server, which is the reason this repository blocks multipart outright where it is not needed. Test in staging.

PHP settings

The same lines, with comments, are in php-fpm/pool-php2shell.conf.

Set them where the application cannot override them (php_admin_value and php_admin_flag in the FPM pool or in the Apache configuration), not with ini_set() or .user.ini, and reload PHP-FPM so every worker is recreated. Set process_control_timeout first (next section), or the reload drops the requests in flight.

; PHP-FPM pool
php_admin_value[memory_limit] = -1
php_admin_value[post_max_size] = 8M
php_admin_value[max_input_nesting_level] = 32
php_admin_flag[session.upload_progress.enabled] = off
; PHP 8.5 only
php_admin_flag[fatal_error_backtraces] = off
; workers are recycled after 500 requests instead of never
pm.max_requests = 500
; stronger, costs performance: the worker is recreated after every request
;pm.max_requests = 1
  • memory_limit = -1 removes the limit an attacker could trip on purpose, but an underlying allocation failure can still reach the vulnerable path. Pair it with a container or cgroup limit that kills only the affected worker (see "Memory cap with systemd").
  • Do not use RLIMIT_AS or RLIMIT_DATA (ulimit -v) as a memory cap. They make mmap() fail, which is the failure that triggers the vulnerable path.
  • post_max_size = 8M is for applications without uploads. If uploads are needed PHP suggests 32M, with upload_max_filesize below it.
  • session.upload_progress.enabled = off blocks the confirmed session paths and not other variants. It is also the only mitigation for CVE-2026-115560, which has no release yet, so keep it off after the update unless you use the feature and your code never calls ini_set('session.serialize_handler', ...).
  • pm.max_requests defaults to 0, so a worker and its state can live until the next reload. 500 bounds that at almost no cost. The advisory suggests 1 if performance allows; neither removes the condition.
  • The advisory asks for a lower max_input_nesting_level than the default 64.
  • Remove or fix settings that emit diagnostics before initialization, and fix fatal errors and execution timeouts an attacker can reach. Hiding messages with error_reporting, display_errors or @ does not help. See "Values passed to PHP per request" and "Startup warnings" below.

Verify values from the web SAPI with a temporary phpinfo() page, not from the CLI, then delete it.

Reloading PHP-FPM without dropping requests

With the default process_control_timeout = 0 a reload sends SIGQUIT and then SIGTERM to every worker at once: the requests in flight are cut and nginx answers 502. On a busy site that is hundreds of failed requests for each reload. Give the workers time to finish, in the [global] section of php-fpm.conf, above the include= line (php-fpm/global-php2shell.conf):

[global]
process_control_timeout = 10s

Added at the end of the file, after include=, the line is read as part of the last pool and FPM refuses the configuration with unknown entry 'process_control_timeout'. The reload that applies the setting still runs with the old value, because the master stops the workers before it re-reads its configuration, so do that first reload at a quiet moment. Test before every reload:

php-fpm7.4 -t && systemctl reload php7.4-fpm

Memory cap with systemd

memory_limit = -1 needs a cap outside PHP. With systemd, a drop-in on the FPM unit sets it (systemd/php-fpm-memory.conf):

# /etc/systemd/system/php7.4-fpm.service.d/php2shell-memory.conf
[Service]
MemoryMax=2G
MemorySwapMax=0
MemoryLimit=2G
OOMPolicy=continue
Restart=on-failure
  • Check the current usage first with systemctl show -p MemoryCurrent php7.4-fpm and size the cap above it. A cap below the usage makes the kernel reclaim and kill right away.
  • Apply it without a restart with systemctl set-property --runtime php7.4-fpm.service MemoryMax=2G MemorySwapMax=0 (MemoryLimit=2G on cgroup v1), then systemctl daemon-reload for the drop-in. OOMPolicy= cannot be set with set-property; daemon-reload applies it to the running service.
  • MemoryMax= and MemorySwapMax= are for cgroup v2, MemoryLimit= for cgroup v1. stat -fc %T /sys/fs/cgroup prints cgroup2fs on v2. There, drop MemoryLimit=: systemd 252 and later log it as deprecated.
  • With the default OOMPolicy=stop on cgroup v2, the first time the kernel kills a worker for memory systemd stops the whole service, every pool with it. continue keeps it running and the master starts a new worker. On cgroup v1 systemd does not see OOM kills, and systemd older than 243 (Debian 10, CentOS 8) ignores the line.
  • Without MemorySwapMax=0 the pool goes to swap at the cap and a runaway worker is not killed until swap runs out.
  • The kernel kills one process, normally the largest worker, but the advisory warns that the master can go too. Restart=on-failure brings the service back in that case.

On systemd 252 with cgroup v2, one worker killed under OOMPolicy=stop left the unit failed with oom-kill; under continue only that request got a 502.

Values passed to PHP per request

On FPM, INI values sent by the web server with each request (PHP_VALUE and PHP_ADMIN_VALUE) are applied before the request starts; the advisory for CVE-2026-109619 names them as the FPM path when one of them produces a diagnostic. Pool values and .user.ini do not behave that way. Find them and move the settings into the pool:

nginx -T 2>/dev/null | grep -E 'fastcgi_param\s+PHP_(ADMIN_)?VALUE'
grep -rnE 'PHP_(ADMIN_)?VALUE' /etc/apache2 /etc/httpd 2>/dev/null

With mod_php the same holds for php_value and php_flag in .htaccess and in the Apache configuration. mod_php is also where the full exploit was shown, so make sure it is not loaded even where Apache is stopped and FPM does the work: a later start of Apache brings it back.

apachectl -M 2>/dev/null | grep -i php    # reads the configuration, works with Apache stopped
a2dismod php7.4                           # Debian, Ubuntu

On RHEL and CentOS the module is loaded by /etc/httpd/conf.modules.d/15-php.conf: comment out its LoadModule line.

Startup warnings

The easiest diagnostics to find are the warnings PHP prints each time FPM starts or reloads. A common cause is extension= lines added to php.ini for modules the distribution already loads from conf.d:

PHP Warning:  PHP Startup: Unable to load dynamic library 'mysqli' (... undefined symbol: mysqlnd_global_stats)
PHP Warning:  Module 'curl' already loaded

mysqli listed in php.ini loads before mysqlnd from conf.d and fails, and curl is loaded twice. These come from module startup, not from a request, so they are not the path the advisory describes. On a production server they came with short bursts of HTTP 500 (HeadersAlreadySentException from the framework) right after a reload, and they bury the warnings that matter. Remove the duplicates and check that FPM starts clean:

php-fpm7.4 -m 2>&1 | grep -ciE 'warning|unable|already loaded'    # expect 0

ISPConfig

ISPConfig rewrites pool files and vhosts from its templates each time a site is saved in the panel, so lines added by hand to pool.d/web*.conf or to the vhost disappear. Put them where the panel keeps them:

  • PHP settings in the site's Options tab, "Custom php.ini settings". For PHP-FPM, ISPConfig writes them into the pool as php_admin_value and php_admin_flag.
  • nginx rules in the same tab, "nginx Directives".
  • Settings for every site in a copy of the template under /usr/local/ispconfig/server/conf-custom/, such as php_fpm_pool.conf.master or nginx_vhost.conf.master. Compare it with the new template after each ISPConfig update.

The [global] section of php-fpm.conf and the systemd drop-in are not managed by ISPConfig.

What's here

  • nginx/php2shell-maps.conf: map blocks for http{}, flags multipart POST, PUT and PATCH minus an upload-route allowlist
  • nginx/php2shell-server-snippet.conf: the if that returns 403, for server{}
  • apache/.htaccess: mod_rewrite rule that returns 403 on multipart POST, PUT and PATCH
  • apache/modsecurity-php2shell.conf: ModSecurity rules on the PHP_SESSION_UPLOAD_PROGRESS field and on malformed bodies, plus optional body limits and a multipart block
  • apache/auth-before-php.conf.example: IP allow-list or basic auth and request limits for Apache
  • nginx/auth-before-php.conf.example: IP allow-list or basic auth and body limit for Nginx
  • cloudflare/custom-rule.md: custom-rule expression, scoped, and what not to do
  • caddy/php2shell.caddyfile: matcher that returns 403 on multipart except upload routes
  • frankenphp/Caddyfile-snippet: disables session upload progress
  • lighttpd/php2shell.conf: multipart block, to scope per vhost
  • litespeed/php2shell.conf: the same rewrite rule for LiteSpeed Enterprise and OpenLiteSpeed, plus where the lsphp php.ini lives
  • php-fpm/99-php2shell.ini: session hardening for php.ini
  • php-fpm/pool-php2shell.conf: the advisory settings as php_admin_* lines for an FPM pool, with pm.max_requests
  • php-fpm/global-php2shell.conf: process_control_timeout, so a reload does not drop requests
  • systemd/php-fpm-memory.conf: drop-in with the memory cap and OOMPolicy=continue
  • test/audit-server.sh: read-only audit of a PHP-FPM server
  • test/verify.sh: checks a running site, any web server, against the multipart block
  • test/check-sync.py: fails when the maps in the container test drift from the shipped ones
  • test/: decision-table check and a self-contained Nginx container test
  • .github/workflows/test.yml: runs all of the above on every push

Why multipart

The advisory lists multipart requests as a way to exhaust memory for CVE-2026-109618, and session upload progress, which PHP processes when a multipart POST carries the field named in session.upload_progress.name (PHP_SESSION_UPLOAD_PROGRESS by default), as a confirmed path for CVE-2026-109619. The parser runs before the application script and before any login. Blocking multipart at the edge where it is not needed removes those paths. It does not cover an application-specific out-of-memory condition caused by other input.

Why the regex catches the comma variant

PHP lowercases the Content-Type and truncates it at the first ;, , or space before it matches a post handler, then looks for boundary anywhere in the header. So Content-Type: multipart/form-data,boundary=x is multipart to PHP, and a rule that only matches multipart/form-data followed by ; or whitespace is bypassable with a comma. Sources (PHP 8.4.26):

Limits

  • Per-route exceptions do not protect those routes. The multipart parser runs before your application code and before authentication, so any route you allow stays exposed.
  • The exploit was demonstrated end to end on mod_php only. For FPM, FrankenPHP and LSAPI the first issue is source analysis, and "not demonstrated" does not mean "not exploitable".
  • PHP gives no list of affected versions below 7.4. Treat PHP 5.x and 7.3 as exposed.
  • .htaccess can be silently ignored: insufficient AllowOverride, ProxyPassMatch or SetHandler proxy:fcgi setups, nested .htaccess with their own RewriteEngine, or a rule placed after an earlier [L]. Use %{HTTP:Content-Type}, not %{CONTENT_TYPE}.
  • From PHP 8.4 request_parse_body() parses multipart on PUT and PATCH too, so the rules include them.
  • Nginx with an error_page cycle: with error_page 403 ... and recursive_error_pages on (the ISPConfig default) a plain if ... return 403; turns into HTTP 500 through an internal redirect. The $pp_errpage map keyed on $uri, plus volatile, prevents it. Verified in production and on an isolated nginx 1.24.
  • A WAF parses multipart differently from PHP. The ModSecurity rule is a layer, not a fix.
  • An empty error log does not prove the server is safe.

Check that it works

After applying a rule, run the checker against staging. It sends empty POST and PUT requests and expects multipart (including the comma and upper-case variants) to be blocked, while GET, urlencoded and JSON requests pass.

bash test/verify.sh https://staging.example.test /upload

The second argument is an upload route that must keep working; leave it out if there is none. A blocked request answers 403, 400 or drops the connection (nginx return 444).

Audit a server

test/audit-server.sh runs on the server, as root, and changes nothing. It reads the effective configuration with php-fpm -tt, nginx -T, apachectl -M and systemctl show, for every PHP-FPM it finds (distribution packages, ISPConfig builds in /opt/php-*, Remi), and checks:

  • memory_limit = -1, session.upload_progress.enabled off and pm.max_requests above 0 as php_admin_* in every pool
  • process_control_timeout above 0
  • no startup warnings from php-fpm -m
  • no RLIMIT_AS or RLIMIT_DATA on the running master and workers, and no LimitAS= or LimitDATA= on the unit
  • MemoryMax= (MemoryLimit= on cgroup v1) and OOMPolicy=continue on the unit
  • no PHP_VALUE or PHP_ADMIN_VALUE passed by nginx or Apache
  • mod_php not loaded or enabled, and php_value in .htaccess when it is
  • no client_max_body_size 0 in the nginx configuration
sudo bash test/audit-server.sh

Each line is PASS, WARN or SKIP, and the exit code is 1 if there is a WARN. The output names files, pools and settings of the server, so read it before you share it.

Test

# decision table, no server needed
python3 test/test_logic.py

# the container test and the shipped maps still agree
python3 test/check-sync.py

# full Nginx container test (needs Docker)
bash test/run-docker-test.sh

The Apache, Caddy, lighttpd and LiteSpeed rules have no container test yet: check them with test/verify.sh.

Every push runs the decision table, the shell syntax check, the drift check and the Nginx container test through .github/workflows/test.yml.

After the update

Update PHP to a patched release and remove these rules where you can. Keep them only for branches that will not be patched, and re-check against the advisory.

Keep session.upload_progress.enabled = off even on a patched release until the bugfix release that fixes CVE-2026-115560 is out: the versions of 21 October do not contain that fix.

Detection

Scanning is worth assuming. As of 23 October 2026 watchTowr reports exploitation attempts against its honeypots since 21 October, Amazon describes the flaws as broadly exploited and traced an attempt back to 17 October, before disclosure, and Cloudflare reports attack traffic against its customers. Huntress and CISA saw none on 21 October, no compromise has been confirmed publicly, and neither CVE is in the CISA KEV catalog yet. Check the live catalog rather than this line.

  • PHP worker crashes (PHP-FPM or LiteSpeed's lsphp) with SIGSEGV or SIGABRT in the logs, especially close to multipart POSTs.
  • Access-log multipart POSTs carrying a PHP_SESSION_UPLOAD_PROGRESS field on routes that never accept uploads.
  • Session files in session.save_path with upload-progress entries for sessions that never uploaded, or rewritten unusually often.
  • Enable log_errors, a working error_log and error_reporting = E_ALL in an isolated environment to find fatal errors and timeouts reachable from outside.

Read more on pasqualepillitteri.it

License

MIT, see LICENSE.

About

Mitigations for PHP installs that cannot update yet: CVE-2026-109618 / CVE-2026-109619 (php2shell) and CVE-2026-115560. Edge rules, auth before PHP, request limits, PHP settings. Unofficial. The real fix is updating PHP.

Topics

Resources

Stars

9 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages