PHP November Security Release: 18 CVEs, PHP-FPM Request Smuggling CVE-2026-129289 and CVE-2026-115560

Introduction

Four weeks after the emergency PHP2Shell patches, PHP shipped another security release on November 19: 8.5.13, 8.4.28, 8.3.37 and 8.2.36, fixing 18 CVEs across every supported branch. The PHP Foundation rates the worst of them critical, though it says they need non-default or uncommon setups. The two to understand are CVE-2026-129289, a CVSS 9.2 request-smuggling bug in PHP-FPM, and CVE-2026-115560, the session serializer RCE that PHP warned about in October but did not fix then. Nothing in this batch is known to be exploited yet. Update anyway, because PHP2Shell went from patch to KEV in five days.

What Happened

PHP pre-announced the release on November 18 without CVE details, then published the fixes on November 19 between 14:00 and 14:30 UTC. PHP 8.6.0 reached general availability the same day.

CVE-2026-129289 (PHP-FPM, CVSS 9.2). If an application opens php://input and does not read the whole body, the unread bytes can be treated as the start of the next FastCGI request on a reused backend connection. That gives an attacker request smuggling into another user's request. It only works when the front end keeps FastCGI connections open between requests:

  • nginx with fastcgi_keep_conn on
  • Apache mod_proxy_fcgi with enablereuse=on
  • HAProxy FastCGI backends that retain connections
  • LiteSpeed LSAPI

Standard nginx and Apache configs that close the FastCGI connection after each request are not affected by the demonstrated attack.

CVE-2026-115560 (session, RCE). PHP published an informational advisory on October 23, said the fix would come in the next regular release, and stated that the October 21 PHP2Shell builds (8.5.12 / 8.4.27 / 8.3.36 / 8.2.35) do not fix it. A site is exposed only when all three of these hold:

  1. session.serialize_handler is set to php_binary or php_serialize in config,
  2. code switches it to php with ini_set() before session_start(), and
  3. session upload progress is enabled, which is the default.

The PHP Foundation had already said this combination is very unlikely in practice.

The rest of the batch, per Aikido's breakdown of the changelog:

CVE Component Issue
CVE-2026-126714 Session Partially overwritten session files → potential RCE
CVE-2026-129481 Session Use-after-free after interrupted session shutdown
CVE-2026-129474 Core Use-after-free in magic method backtraces → potential RCE
CVE-2026-129182 Core Use-after-free repopulating fatal error backtrace after shutdown
CVE-2026-129496 Date Timezone cache use-after-free → potential RCE
CVE-2026-129501 DOM Double free during XML serialization
CVE-2026-129538 CGI Out-of-bounds read/write in PATH_INFO correction
CVE-2026-129291 FPM epoll use-after-free → local privilege escalation
CVE-2026-129292 FPM XSS on the status page via unescaped fields
CVE-2026-128090 Phar OpenSSL signature bypass via algorithm downgrade
CVE-2026-129506 SOAP Use-after-free on DTD entity references
CVE-2026-15353 SOAP Information disclosure via malformed array dimensions
CVE-2026-128064 Standard Mixed-case HTTPS:// / FTPS:// URLs can fall back to plaintext
CVE-2026-127960 Standard DoS via folded HTTP response headers
CVE-2026-129438 URI Incorrect RFC 3986 serialization for oversized input
CVE-2026-48841 URI Bundled uriparser updated to 1.0.3

PHP has published severity ratings only for the headline issues, and outlets disagree on the count of "critical" bugs, from seven up to the full list. Go by the release and the changelog rather than any single tally.

Why It Matters

PHP2Shell showed how fast a PHP memory-safety disclosure turns into real attacks. Five days after the October release it was in CISA KEV. Within weeks, attackers had used it to root Palo Alto and Check Point appliances that bundle PHP. Several bugs in this batch are in the same family: use-after-free and double-free in core, session, date, DOM and SOAP, with "potential RCE" attached to four of them. Expect researchers and attackers to diff 8.5.12 against 8.5.13 straight away.

CVE-2026-129289 is the one to watch on modern stacks. FastCGI keep-alive is a common performance tweak, and request smuggling on a shared PHP-FPM pool means one client can inject requests into another user's session. In practice that means account takeover, cache poisoning and bypassing front-end access controls, all without any memory corruption. The Phar signature bypass also matters for anyone who verifies signed .phar updates, such as Composer-style self-updaters and plugin installers.

Who Is Affected

  • Every PHP release before 8.5.13, 8.4.28, 8.3.37 and 8.2.36. That includes the October PHP2Shell builds.
  • PHP 8.1, 8.0, 7.x and 5.x: no upstream fix. Treat them as affected and look to your distro or LTS vendor for backports. Backported packages often keep the old version string, so check the changelog for the CVE IDs.
  • CVE-2026-129289: PHP-FPM or LSAPI behind a front end that reuses FastCGI connections.
  • CVE-2026-115560: apps that switch session.serialize_handler at runtime, with upload progress on.
  • CVE-2026-129538: php-cgi deployments.
  • PHP 8.2 reaches end of life on December 31, 2026. Plan the move to 8.4 or 8.5 now.
  • Appliances and panels that bundle PHP have to ship their own update.

How to Protect Yourself

1. Inventory PHP versions on every web SAPI

php -v
for b in /usr/sbin/php-fpm* /usr/bin/php-cgi*; do [ -x "$b" ] && echo "$b: $($b -v | head -1)"; done
apachectl -M 2>/dev/null | grep -i php

# containers
docker ps --format '{{.Names}}' | while read n; do echo "$n: $(docker exec "$n" php -r 'echo PHP_VERSION;' 2>/dev/null)"; done

2. Update and restart workers

# Debian / Ubuntu
sudo apt update && sudo apt install --only-upgrade 'php*'
sudo systemctl restart 'php*-fpm' apache2 2>/dev/null

# RHEL / Alma / Rocky
sudo dnf upgrade 'php*'
sudo systemctl restart php-fpm httpd

# Docker: rebuild on php:8.5.13-*, php:8.4.28-*, php:8.3.37-*, php:8.2.36-*
grep -rnE '^FROM .*php:' --include=Dockerfile* .

3. CVE-2026-129289: check for FastCGI connection reuse

# nginx
nginx -T 2>/dev/null | grep -n 'fastcgi_keep_conn'
# Apache
grep -rnE 'enablereuse=on' /etc/apache2 /etc/httpd 2>/dev/null
# HAProxy
grep -nE 'use-fcgi-app|fcgi-app' /etc/haproxy/haproxy.cfg 2>/dev/null

Until you are patched, turn reuse off (fastcgi_keep_conn off;, or remove enablereuse=on) and reload. PHP says that reading php://input to the end in application code is not a reliable fix on its own. Never expose PHP-FPM or LSAPI listeners to untrusted networks:

ss -ltnp | grep -E 'php-fpm|lsphp'   # should be 127.0.0.1 or a unix socket

4. CVE-2026-115560: look for runtime serializer switching

grep -rnE "ini_set\(\s*['\"]session\.serialize_handler" --include='*.php' /var/www 2>/dev/null
grep -rnE "['\"]serialize_handler['\"]\s*=>" --include='*.php' /var/www 2>/dev/null

If you find matches and cannot patch yet, disable upload progress at the server level. ini_set(), .user.ini and .htaccess are too late, because the directive is INI_PERDIR:

; php.ini or conf.d drop-in
session.upload_progress.enabled = 0
; PHP-FPM pool
php_admin_flag[session.upload_progress.enabled] = off

Restart workers and verify the setting through the web SAPI, not php -i on the CLI. If you already applied this setting for PHP2Shell, keep it. For a durable fix, pick one serializer in config and pin it with php_admin_value[session.serialize_handler].

5. Watch for the next round of crashes

Failed exploitation of the memory-safety bugs looks the same as it did with PHP2Shell: workers dying on signal 11 or 6.

sudo grep -hE 'child [0-9]+ exited on signal (11|6)|SIGSEGV' /var/log/php*-fpm.log /var/log/apache2/error.log /var/log/httpd/error_log 2>/dev/null | tail
ps -eo pid,ppid,user,cmd --forest | grep -A2 -E '[p]hp-fpm|[a]pache2|[h]ttpd' | grep -E 'sh -c|bash|curl|wget|python|nc '

Source