CVE monitoring and patch operations instead of finding out by chance - WZ-IT records where PHP runs in your environment and how the front end is connected, applies updates and restarts the services. Book an initial consultation · Server audit · Nextcloud operations
On 19 November 2026 the PHP project released PHP 8.5.13, 8.4.28, 8.3.37 and 8.2.36. The ChangeLog lists 18 CVE identifiers, seven of them critical. Many reports focus on CVE-2026-129289, a request smuggling flaw in PHP-FPM that can lead to code execution when FastCGI connections are kept open. Alongside it are three session flaws, two of which are connected to the session.upload_progress setting.
This article does not repeat what applies to the flaws of 21 October 2026 (PHP2Shell, CVE-2026-109618 and CVE-2026-109619). It covers the question that decides the effort this time: which configuration makes the PHP-FPM flaw exploitable, and what do common open-source applications and their images ship for it? For each application it names the documented PHP state, explains the session flaws and describes why PHP-FPM has to restart after the update.
Table of contents
- In short
- What was released on 19 November 2026
- When the PHP-FPM flaw applies
- The session flaws and upload progress
- Which applications and images are affected
- How to check your version
- Immediate measures and workarounds
- Why PHP-FPM has to be restarted
- Update paths by deployment type
- Is the flaw being exploited?
- Timeline
- Common misconceptions
- Status and methodology
- Our approach at WZ-IT
- Further guides
In short
| Question | Answer |
|---|---|
| Release | 19 November 2026: PHP 8.5.13, 8.4.28, 8.3.37 and 8.2.36 (php.net), together with PHP 8.6.0 |
| Scope | 18 CVE identifiers in the ChangeLog of 8.5.13, 17 of them with a GitHub advisory (7 critical, 4 high, 6 medium); 8.4.28: 14, 8.3.37: 13, 8.2.36: 12 |
| PHP-FPM flaw | CVE-2026-129289: request smuggling when FastCGI connections are kept open and an application does not read php://input to the end; nginx and Apache do not keep connections open by default |
| Session flaws | CVE-2026-115560 (serializer mismatch with upload progress enabled), CVE-2026-126714 (files handler), CVE-2026-129481 (PHP 8.3 and later) |
| Severity | CVSS 4.0: 9.2 for all seven critical flaws (assigned by the PHP project as CNA); for CVE-2026-129289 Red Hat CVSS 3.1: 8.1, Ubuntu priority Medium |
| New default | session.upload_progress.enabled is 0 in the source code of the new releases; the PHP manual still says 1 |
| Update | 8.5.13, 8.4.28, 8.3.37 or 8.2.36, then restart PHP-FPM (Apache with mod_php), pull container images again |
| Exploitation | As of 20 November 2026, 17:48 UTC, none of the November CVEs in the CISA KEV catalog; CISA assessment in the CVE record: Exploitation none |
What was released on 19 November 2026
On 18 November 2026 the PHP project announced security releases for 19 November between 14:00 and 14:30 UTC. They were part of the regular four-week cycle and would affect all supported branches; the highest severity was critical, and the critical cases required non-default configurations or uncommon setups (php.net). The GitHub advisories appeared on 19 November. The release announcements name no CVEs; details are in the ChangeLog and in the advisories of the repository php/php-src.
The seven critical flaws, each CVSS 4.0 9.2:
| CVE | Component | Core | Affected versions | Condition according to the advisory |
|---|---|---|---|---|
| CVE-2026-129289 | PHP-FPM | Request smuggling through an unread request body | from 5.6.0 up to before 8.2.36, 8.3.37, 8.4.28, 8.5.13 | Front end keeps FastCGI connections open; application opens php://input without reading it to the end |
| CVE-2026-115560 | Session | Serializer mismatch, code execution with a matching gadget chain | from 5.4.0 (variant with php_serialize from 5.5.4) up to before the fixed versions |
Upload progress enabled; writer with php_binary or php_serialize, reader with php |
| CVE-2026-126714 | Session | Partially overwritten session files | from 4.1.0 up to before the fixed versions | files handler (default), influence on session data, interrupted write, gadget chain |
| CVE-2026-129496 | Date | Use-after-free in the timezone cache | 8.2 to 8.5, each before its fixed version; according to the advisory, based on code inspection, the code has been in PHP since 5.3.18 and 5.4.8 respectively (an earlier form in 5.2.10, 5.2.12 to 5.2.17 and 5.3.0 to 5.3.8), and there is no fixed version for unsupported branches | Startup diagnostic in the file-based error_log and an early request error in a fresh worker |
| CVE-2026-129481 | Session | Use-after-free after an interrupted session shutdown | 8.3 and later only | Fatal error in a callback during session cleanup, later access by the same worker |
| CVE-2026-129474 | Core | Use-after-free in backtraces of magic methods | 8.5 only | fatal_error_backtraces=1 and zend.exception_ignore_args=0 |
| CVE-2026-129182 | Core | Use-after-free of the fatal error backtrace after shutdown | 8.5 only | Specific shutdown configuration; incomplete fix for CVE-2026-109619 |
Further flaws according to the ChangeLog: CVE-2026-129291 (PHP-FPM, local privilege escalation, CVSS 4.0 8.8), CVE-2026-128064, CVE-2026-127960, CVE-2026-128090 (high) as well as CVE-2026-129438, CVE-2026-129538, CVE-2026-129506, CVE-2026-129501, CVE-2026-129292 and CVE-2026-15353 (medium). CVE-2026-48841 is the update of the bundled uriparser library to 1.0.3. Which branch fixes which CVEs is listed in the ChangeLog; CVE-2026-129474 and CVE-2026-129182 affect only PHP 8.5.
When the PHP-FPM flaw applies
CVE-2026-129289 arises because at the end of a request PHP does not discard the request body when the application has opened php://input without reading it to the end. If the front end keeps the FastCGI connection to PHP-FPM open, PHP-FPM interprets the leftover bytes as a new FastCGI request. In it the attacker can set parameters such as SCRIPT_FILENAME, REMOTE_ADDR and PHP_ADMIN_VALUE and, if the path of an existing PHP script is known, set allow_url_include=1 and auto_prepend_file=php://input, so that the content of the attacker's own request runs as PHP code (GHSA-qc49-hpvq-g2w6). The advisory names neither memory corruption nor race conditions as a prerequisite. According to code inspection the affected code has been present since PHP 5.6.0, so PHP from 5.6.0 up to before the fixed versions is affected; for older versions the advisory does not state that they are affected.
Three conditions have to be met at the same time:
- PHP-FPM with a version from 5.6.0 up to before 8.2.36, 8.3.37, 8.4.28 or 8.5.13 respectively. For LSAPI the advisory establishes only the disruption of subsequent requests, not code execution (see below). mod_php does not use FastCGI and is not affected by this flaw.
- A front end that keeps backend connections open. The advisory names nginx with
fastcgi_keep_conn on, Apachemod_proxy_fcgiwithenablereuse=on, HAProxy FastCGI applications that retain backend connections, and LSAPI with a persistent connection; for LSAPI it establishes only the disruption of subsequent requests, not code execution. Standard configurations that close the connection after each request are not affected by the demonstrated attack. - An application path that opens
php://inputwithout reading it to the end. The advisory mentions frameworks that create the input stream before routing or authentication. Which applications do this is not documented here.
What the front ends do by default:
| Front end | Default | When affected | Source |
|---|---|---|---|
| nginx | fastcgi_keep_conn off: the FastCGI server closes the connection after the response |
With fastcgi_keep_conn on |
nginx documentation |
Apache with mod_proxy_fcgi |
Reuse off: the connection is not held open after the request | With enablereuse=on |
Apache documentation |
Caddy with php_fastcgi |
New connection per request, keep-alive is marked as a TODO in the source code | Not according to the source code, main branch | Caddy source code |
| HAProxy with FastCGI | Not checked | When backend connections are retained | GHSA-qc49-hpvq-g2w6 |
| LSAPI (LiteSpeed) | Persistent connection | Requests on the connection can be disrupted; the advisory does not establish code execution through LSAPI | GHSA-qc49-hpvq-g2w6 |
The nginx default files fastcgi_params and fastcgi.conf contain no fastcgi_keep_conn (nginx repository, as of 18 November 2026). Whoever does not set the directive does not have it. Whoever set it for performance reasons finds it like this:
# nginx: search the effective configuration for keep-alive to FastCGI
nginx -T 2>/dev/null | grep -n -i -e fastcgi_keep_conn -e 'keepalive '
# Apache: is mod_proxy_fcgi loaded, and is enablereuse set?
apachectl -t -D DUMP_MODULES 2>/dev/null | grep -i -e proxy_fcgi -e php
grep -rn -i enablereuse /etc/apache2 /etc/httpd 2>/dev/null
nginx -T prints the configuration including included files (nginx documentation), -D DUMP_MODULES lists the loaded Apache modules (Apache documentation). A keepalive in the upstream block is a reason to look at the FastCGI settings more closely. The advisory also warns: PHP-FPM and LSAPI listeners do not belong on untrusted networks.
The session flaws and upload progress
Three flaws are in the session extension. They have different triggers and react differently to session.upload_progress.enabled=0:
| CVE | Trigger | Does session.upload_progress.enabled=0 help? |
Interim measure without update |
|---|---|---|---|
| CVE-2026-115560 | Upload progress enabled; one context writes with php_binary or php_serialize, another reads the same session with php; the reading application needs a gadget chain |
Yes, it removes the upload progress path described in the advisory | Switch the setting off; if the feature stays, use the same session.serialize_handler everywhere |
| CVE-2026-126714 | files handler (PHP default): an interrupted write leaves a mixture of the new and old session file; attackers must be able to influence session content, the application needs a gadget chain | Partly: removes the input path through upload progress, not applications that store attacker data in sessions themselves | Switch the setting off, limit the amount of untrusted data in sessions, use session storage that writes atomically |
| CVE-2026-129481 | Fatal error in a callback during session cleanup (for example in the stream_close() of a stream wrapper), later the same worker; PHP 8.3 and later only |
No, according to the advisory without effect | pm.max_requests=1 in the FPM pool, with a performance cost |
Sources: GHSA-658j-r8rw-q6mx, GHSA-fm2m-fjf5-v3cp, GHSA-g3vh-838c-65v7.
How CVE-2026-115560 comes about. If a multipart request appends the value PHP_SESSION_UPLOAD_PROGRESS, PHP writes it into a session key. The php format separates key and value with |. If a reader using php reads an entry that a writer using php_binary or php_serialize stored, a delimiter set by the attacker can cause bytes from the key to be deserialized. The mismatch arises when an application sets session.serialize_handler to php before session_start(), or between separate applications, pools, virtual hosts and CLI processes with different settings. Separate FPM pools do not automatically separate the session storage (GHSA-658j-r8rw-q6mx). Upload progress has existed since PHP 5.4.0, and the variant with a php_serialize writer requires PHP 5.5.4 or later (code inspection according to the advisory); older, unsupported releases may therefore also be affected. On 23 October 2026 the PHP project had advised in advance to set session.upload_progress.enabled=0 and classified the flaw as an "unlikely RCE vector": applications that never switch the serializer at runtime could leave the feature enabled (php.net).
The default has changed. The fix disables upload progress by default and rejects runtime changes to session settings while it is active (GHSA-658j-r8rw-q6mx). In the source code the default in ext/session/session.c is 1 in 8.5.12 and 8.2.35 and 0 in 8.5.13, 8.4.28, 8.3.37, 8.2.36 and 8.6.0 (8.5.13, 8.5.12). The sources contradict each other here: on 21 November 2026 the PHP manual still says "Defaults to 1", and the php.ini-production template of 8.5.13 lists ;session.upload_progress.enabled = On with "Default Value: On". An explicitly set value overrides the new default (Red Hat). Where session.upload_progress.enabled=1 is set in an old php.ini, a pool or an image, the feature stays on after the update.
Side effect for applications that change session settings at runtime. If upload progress is active, PHP 8.5.13 throws an error on runtime changes to session settings such as session.serialize_handler ("Session ini settings cannot be changed at runtime when session.upload_progress.enabled is On", source code). This touches applications that do exactly that: Moodle and Roundcube set session.serialize_handler to php via ini_set(), Matomo to php_serialize. With the new default "off" this does not occur; it affects only installations that explicitly switch upload progress on.
Which applications and images are affected
The table lists only statements that are documented in the project's repository, Dockerfile, registry configuration, documentation or tracker. "Can be affected" means for every row: the image can contain at least one of the flaws that apply to its PHP branch as long as PHP is below 8.5.13, 8.4.28, 8.3.37 or 8.2.36 (Session, Date and others). Not every flaw applies to every version and every configuration: CVE-2026-129474 and CVE-2026-129182 affect only PHP 8.5, CVE-2026-129481 PHP 8.3 and later, and several flaws need a matching configuration or a gadget chain. CVE-2026-129289 additionally needs FPM with a FastCGI connection kept open; in the checked example configurations and standard setups of the projects this is not set. Whether an application opens php://input without reading it fully has not been checked. The image status applies as of 21 November 2026 and can change at short notice with every rebuild. The dates are the creation dates from the image configuration in the registry, not necessarily the time from which a tag was available there.
| Application or delivery | How PHP gets in | Can be affected? | Corrected state and action | Source |
|---|---|---|---|---|
| Official PHP image (base of many images) | php:<branch>-apache is mod_php (prefork, --with-apxs2); php:<branch>-fpm is FPM without a front end and sets catch_workers_output = yes |
Older tags yes. Apache variant: CVE-2026-129289 not applicable. FPM variant: depends on your own front end; CVE-2026-129291 (local) relates to pools with catch_workers_output |
Tags with 8.5.13, 8.4.28, 8.3.37, 8.2.36, images of 19 Nov; pull the image again | Dockerfile apache, Dockerfile fpm, GHSA-jm4m-qfqx-967m |
| Nextcloud, official Docker image (apache, fpm, fpm-alpine) | Base php:8.5-apache-trixie or -fpm (Nextcloud 35 and 34), 8.4 (33), 8.3 (32) |
Older images yes. apache: mod_php. fpm: your own front end; the project's example nginx.conf contains no fastcgi_keep_conn |
Images of 19 Nov: PHP 8.5.13 (35, 34), 8.4.28 (33), 8.3.37 (32); since 23 Oct the image sets session.upload_progress.enabled=0 and uses igbinary as session serializer; pull the image again |
nextcloud/docker, commit 7e9e152f, nginx.conf example |
| Nextcloud All-in-One | php:8.4-fpm-alpine3.24 container behind an Apache container (mod_proxy_fcgi, no enablereuse); sessions in Redis |
Older images yes. Keep-alive prerequisite not met in the default configuration. CVE-2026-126714 affects the files handler, not Redis | ghcr.io/nextcloud-releases/aio-nextcloud:latest of 19 Nov: PHP 8.4.28 (predecessor 8.4.27); recreate the containers with the new image |
all-in-one, Apache configuration, GHSA-fm2m-fjf5-v3cp |
| Nextcloud snap | PHP 8.4.28 is built into the snap from source; Apache with PHP-FPM over a Unix socket (mod_proxy_fcgi, no enablereuse) |
Older snaps yes | Snap Store, stable since 19 Nov: 35.0.3snap2, 34.0.6snap2, 33.0.11snap2 with 8.4.28. Track 32/stable shows 32.0.15snap1 of 24 Sep with PHP 8.3.32, not fixed as of 21 Nov | nextcloud-snap, Snap Store |
| WordPress, official Docker image | php:*-apache (mod_php) or php:*-fpm |
Older images yes; no keep-alive setting in the repository, the WordPress documentation for nginx names fastcgi_pass without fastcgi_keep_conn |
wordpress:latest and :fpm PHP 8.3.37, :php8.4-fpm 8.4.28, :php8.5-apache 8.5.13 (19 Nov); pull the image again |
docker-library/wordpress, WordPress documentation |
| Matomo, Docker image | php:8.4-apache, -fpm, -fpm-alpine |
Older images yes; nginx example without fastcgi_keep_conn. Matomo sets session.serialize_handler=php_serialize at runtime |
matomo:latest, :fpm, :fpm-alpine with PHP 8.4.28 (19 Nov); pull the image again |
matomo-org/docker, Session.php |
| Roundcube, Docker image | php:8.4-apache and -fpm |
Older images yes; examples without fastcgi_keep_conn. The application sets session.serialize_handler=php at runtime |
Version 1.7.6 with PHP 8.4.28 (20 Nov); since 26 Oct the image sets session.upload_progress.enabled=Off; pull the image again |
roundcubemail-docker, commit 8737b949 |
| Kimai, Docker image | php:8.5-apache or php:8.5-fpm-alpine |
Older images yes | Version 2.72.1 with PHP 8.5.13 (20 Nov); the image copies php.ini-production and since 11 Nov sets session.upload_progress.enabled = Off and fatal_error_backtraces = Off |
Dockerfile, commit 8f9df672 |
| Mautic, Docker image | php:8.3-apache-bookworm by default, flavour apache or fpm |
Older images yes; the fpm-nginx example contains no fastcgi_keep_conn |
mautic/mautic:latest, :7-apache, :6-apache, :7-fpm with PHP 8.3.37 (19 Nov) |
docker-mautic |
| Snipe-IT, Docker image | Ubuntu 26.04 with libapache2-mod-php8.5 from the Ubuntu packages (mod_php) |
Older images yes. The Dockerfile sets session.upload_progress.enabled = 0 only in the CLI php.ini (since 20 Nov), not in the Apache php.ini |
snipe/snipe-it:latest of 20 Nov contains PHP 8.5.4-0ubuntu1.6, the fixed Ubuntu version; Alpine and FPM variants not checked |
Dockerfile, Ubuntu tracker |
| BookStack, LinuxServer.io image | Alpine packages php85 and php85-fpm, nginx in the same container (FastCGI on 127.0.0.1:9000, no fastcgi_keep_conn) |
Older images yes. BookStack ships no image of its own, the documentation points to community images | linuxserver/bookstack v26.09.3-ls300 of 20 Nov with PHP 8.5.13-r0 |
BookStack documentation, docker-bookstack, base image |
| BookStack, Ubuntu 26.04 script | System packages php-fpm behind Apache (proxy_fcgi), neither enablereuse nor nginx |
Older package states yes | Ubuntu 26.04: php8.5 8.5.4-0ubuntu1.6 since 20 Nov; update packages, restart PHP-FPM |
Installation script, Ubuntu tracker |
| GLPI, official image | Apache with PHP (php:apache, mod_php), php.ini-production |
Yes: glpi/glpi:latest (12.0.2) and :11.0.13 contain PHP 8.5.12 |
Not fixed as of 21 Nov; the nightly tags (created on 21 Nov) contain 8.5.13, the project rebuilds release images regularly; pull again after the rebuild | docker-images, Docker Hub |
Moodle, moodlehq/moodle-php-apache |
Apache image php:*-apache, according to the README for Moodle development |
Tag 8.4 yes (PHP 8.4.27, built on 22 Oct); tags 8.2, 8.3, 8.5, 8.6 with the corrected version. Moodle sets session.serialize_handler=php at runtime |
Pull tag 8.4 again once it has been rebuilt |
moodle-php-apache, Moodle source code |
For Moodle the documentation on moodle.org could not be retrieved at the time of checking because of bot protection; the row is limited to the development image and the source code.
How to check your version
What counts is the version the web server actually serves. The command line can have a different PHP version than PHP-FPM or mod_php.
# Command line and PHP-FPM (binary name depends on the distribution, e.g. php-fpm8.4)
php -v
php-fpm -v
# Effective settings of PHP-FPM
php-fpm -i | grep -i -e 'session.upload_progress.enabled' -e 'session.serialize_handler'
php-fpm -tt | grep -i -e catch_workers_output -e php_admin -e pm.max_requests
# Is the process still running the old program?
ps -eo pid,lstart,args | grep '[p]hp-fpm: master'
ls -l /proc/$(pgrep -o -f 'php-fpm: master')/exe
php-fpm -v prints the version, -i PHP information and configuration, -tt tests the configuration and dumps it (php-fpm man page). If the program file was replaced during the update, the link in /proc/<PID>/exe ends in (deleted) (proc(5)): the process is then still running the old state. If there are several FPM masters, pgrep -o shows only the oldest.
Backport packages. Debian, Ubuntu and Red Hat often ship fixes with an unchanged major version. For 24.04 Ubuntu names php8.3 8.3.6-0ubuntu0.24.04.14 as fixed. The version number from php -v alone therefore does not prove the state; compare it with the distribution's tracker:
# Debian and Ubuntu: installed package versions
dpkg -l 'php*' 'libapache2-mod-php*' | grep '^ii'
# RHEL and derivatives: changelog of the installed package
rpm -q --changelog php-fpm | grep -m3 CVE-2026-129289
Docker and Kubernetes.
# Version in the container and in the image (environment variable of the official php images)
docker exec <container> php -v
docker image inspect <image> --format '{{range .Config.Env}}{{println .}}{{end}}' | grep '^PHP_VERSION='
# Kubernetes
kubectl exec deploy/<name> -- php -v
kubectl get pods -l app=<name> -o jsonpath='{range .items[*]}{.metadata.name}{" "}{.status.containerStatuses[*].imageID}{"\n"}{end}'
Images that are not built on the official PHP image (Ubuntu packages for Snipe-IT, Alpine packages for LinuxServer) do not set PHP_VERSION; there php -v in the container shows the version.
Immediate measures and workarounds
The PHP project names the update as the actual solution. The following measures come from the advisories and the Red Hat bulletin RHSB-2026-020. Changes to the PHP or FPM configuration take effect only after a reload or restart of PHP-FPM (master and workers), or of Apache with mod_php; they apply per FPM pool, virtual host and application override, and changing only the CLI configuration is not enough. Changes to fastcgi_keep_conn (nginx) or enablereuse (Apache), by contrast, need a reload of the respective web server. Red Hat advises testing them outside production first and removing them after the update.
| Measure | Works against | Limit |
|---|---|---|
Keep-alive to FastCGI off: nginx fastcgi_keep_conn off, Apache enablereuse off, other front ends close after every request |
CVE-2026-129289 | Covers only this path; after an nginx change, test the configuration (nginx -t) and reload nginx, after an Apache change reload Apache. Reading php://input to the end is, according to the advisory, unreliable as the sole measure because error paths can skip it |
session.upload_progress.enabled=0 in php.ini, pool or server configuration |
CVE-2026-115560 (upload path), partly CVE-2026-126714 | Does not work against CVE-2026-129481; according to the advisory ini_set() is too late, and the setting belongs in php.ini, the pool or the server configuration before the request data is processed; the advisory makes no statement on .user.ini; by our reading of the source, PHP-FPM 8.5.13 processes it before reading the POST data (SAPI.h), but do not rely on that; where the feature is needed, use the same session.serialize_handler everywhere |
pm.max_requests=1 in the FPM pool |
CVE-2026-129481 | Red Hat names a significant performance cost; prevents only the demonstrated sequence of two requests in the same worker |
fatal_error_backtraces=0 and zend.exception_ignore_args=1 |
CVE-2026-129474, CVE-2026-129182 (PHP 8.5 only) | fatal_error_backtraces=0 works against both flaws; zend.exception_ignore_args=1 alone does not cover both. display_errors=0 does not prevent the backtrace; check pool and application overrides |
Fix configuration errors and deprecated settings that produce diagnostics at startup, or error_log = syslog; restart master and workers |
CVE-2026-129496 | Changes to $_REQUEST or cookie settings are not a complete remedy |
catch_workers_output = no and the FPM master not running as root |
CVE-2026-129291 (local) | A non-root master limits the privileges but does not fix the flaw |
Where patching is not possible at once, network separation remains: access to web applications is restricted, and FPM listeners are not reachable from outside. Outdated branches (8.1 and older) receive no updates from the PHP project; for CVE-2026-129291 and CVE-2026-127960 the advisory lists 8.1 as affected without a patch.
Why PHP-FPM has to be restarted
A package update replaces files on disk. Running processes keep the program code they loaded at startup. With PHP-FPM, the master process creates the workers by fork from its own memory image; even workers replaced through pm.max_requests therefore take over the master's code as long as it has not been reloaded (our own classification of the FPM process structure, not a quotation from the advisories). The FPM man page describes SIGUSR2 as "graceful reload of all workers + reload of fpm conf/binary" (php-fpm man page). Added to that is the state of the workers: with CVE-2026-129289, changed administrator settings can stay active for later requests of the same worker, and the flaws CVE-2026-129481, CVE-2026-129182 and CVE-2026-129496 rely on state surviving between requests in the same worker. A restart ends those workers.
Which service restarts depends on the deployment type:
| Deployment type | Where the PHP code runs | What to restart | Note |
|---|---|---|---|
| PHP-FPM behind nginx | FPM master and workers | PHP-FPM (restart, or reload with SIGUSR2) |
nginx contains no PHP; nginx needs a test and reload only after a change to its own configuration, such as fastcgi_keep_conn |
PHP-FPM behind Apache (mod_proxy_fcgi) |
FPM master and workers | PHP-FPM | Apache only after configuration changes (reload), for example enablereuse |
| Apache with mod_php | Apache processes | Apache | The PHP project names mod_php explicitly as an affected deployment type |
| Docker and Kubernetes | Container | Recreate the container with the new image, roll out the pods again | An update of the host packages does not change the PHP version in the container |
For RHEL, Red Hat names systemctl restart php-fpm.service. Service names on other distributions differ; systemctl list-units '*fpm*' shows them. Whether a package update restarts the service itself depends on the distribution and package and is not documented here. After the update therefore: check the running version, not only the installed one.
Update paths by deployment type
System packages. Status of the distributions on 21 November 2026 for CVE-2026-129289 and the other flaws of the update:
| Distribution | Fixed package state | Source |
|---|---|---|
| Debian 12 (bookworm) | php8.2 8.2.36-1~deb12u1 (DLA-4916-1) |
Debian tracker |
| Debian 13 (trixie) | php8.4 8.4.28-1~deb13u1 (DSA-6641-1) |
Debian tracker |
| Debian forky and sid | forky 8.4.27-1 listed as vulnerable, sid 8.4.28-3 fixed | Debian tracker |
| Ubuntu 22.04, 24.04 | php8.1 8.1.2-1ubuntu2.29, php8.3 8.3.6-0ubuntu0.24.04.14 |
Ubuntu tracker |
| Ubuntu 26.04 LTS, 26.10 | php8.5 8.5.4-0ubuntu1.6, 8.5.9-0ubuntu2.2 |
Ubuntu tracker |
| Ubuntu 20.04 and older | Through Ubuntu Pro (ESM), own version per release | Ubuntu tracker |
| RHEL 10 | RHSA-2026:100821: php8.4 8.4.28-1.el10_3 (erratum of 20 Nov) |
Security Data API |
| RHEL 7, 8, 9 | As of 21 Nov listed as "Affected" in the Security Data API, no erratum named; the bulletin RHSB-2026-020 lists RHEL 7 to 10 as affected | Security Data API, RHSB-2026-020 |
Ubuntu published the packages in the security pocket on 20 November 2026 (Launchpad). Ubuntu lists CVE-2026-129289 with priority Medium, the CVSS 4.0 value of the PHP project is 9.2, Red Hat lists CVSS 3.1 with 8.1. After apt upgrade or dnf update, restart PHP-FPM or Apache and check the running version.
Docker. The base image contains PHP; a host update changes nothing. Pull the new image, recreate the container, check the version in the container:
docker compose pull && docker compose up -d
docker exec <container> php -v
With mutable tags such as nextcloud:35-apache or php:8.4-fpm, pulling fetches the current state; an image pinned to a digest stays on the old state until the digest is changed. The GLPI project describes pinning by digest in its documentation. Rebuild your own images on a PHP base image even if your own code is unchanged. Whoever wrote session.upload_progress.enabled=1 into a supplied php.ini or the pool configuration removes or reviews the line.
Kubernetes. Change the image tag or digest in the deployment, or roll out again with imagePullPolicy: Always for mutable tags, then kubectl rollout restart deployment/<name>. Afterwards check that all pods use the new image (imageID) and that php -v shows the fixed version. Helm charts that pin a version tag or a digest need the new value in the values file; a mutable tag is resolved again when pods are recreated with imagePullPolicy: Always.
Is the flaw being exploited?
As of 20 November 2026, 17:48 UTC (catalog version 2026.11.20), none of the November CVEs is in the CISA catalog of known exploited vulnerabilities. CVE-2026-109618 and CVE-2026-109619 from October have been in it since 26 October 2026. The CISA assessment (SSVC) in the CVE record of CVE-2026-129289 of 20 November reads: Exploitation none, Automatable yes, Technical Impact total; for CVE-2026-115560 and CVE-2026-129481 "Automatable" is no, for CVE-2026-126714 yes.
Repositories that describe themselves as an exploit or proof of concept for CVE-2026-129289, CVE-2026-126714 and CVE-2026-115560 can be found publicly on GitHub. Whether they work has not been verified, and they are not linked here. The advisory for CVE-2026-115560 documents a reproduction with Guzzle and Monolog gadget chains. The causes are public; whoever runs applications with FastCGI keep-alive or active upload progress should therefore not wait for reports of exploitation.
Timeline
| Date | Event |
|---|---|
| 21 October 2026 | PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35 fix CVE-2026-109618 and CVE-2026-109619 (PHP2Shell) |
| 23 October 2026 | PHP notice: set session.upload_progress.enabled=0; the Nextcloud image adopts the setting the same day |
| 26 October 2026 | CISA adds CVE-2026-109618 and CVE-2026-109619 to the KEV catalog; the Roundcube image switches upload progress off |
| 11 November 2026 | The Kimai image sets session.upload_progress.enabled = Off |
| 18 November 2026 | PHP announces security releases for 19 November |
| 19 November 2026 | PHP 8.5.13, 8.4.28, 8.3.37, 8.2.36 and 8.6.0 appear (14:00 to 14:30 UTC announced); the official php images, the Nextcloud images, Nextcloud All-in-One and the Nextcloud snaps follow the same day |
| 20 November 2026 | Ubuntu publishes fixed packages; Red Hat publishes RHSA-2026:100821 for RHEL 10; Snipe-IT and LinuxServer BookStack rebuild images |
Common misconceptions
| Assumption | Finding |
|---|---|
| "PHP-FPM behind nginx is automatically vulnerable to CVE-2026-129289." | No. It requires connection reuse (fastcgi_keep_conn on, enablereuse=on, HAProxy; for LSAPI the advisory shows only the disruption of subsequent requests). Both web servers do not do this by default (nginx, Apache). |
| "Without keep-alive the update is unnecessary." | No. The flaws in Session, Date and other extensions do not depend on the FastCGI setup. |
| "With mod_php I am fine for this update." | No. Only CVE-2026-129289 affects PHP-FPM and LSAPI. According to the advisory CVE-2026-129496 affects FPM and mod_php. |
"session.upload_progress.enabled=0 protects against all session flaws." |
No. CVE-2026-129481 remains untouched, for CVE-2026-126714 only one input path is removed (GHSA-g3vh-838c-65v7, GHSA-fm2m-fjf5-v3cp). |
| "The new default switches upload progress off, so my system is safe." | Only if neither php.ini nor pool or image explicitly switch it on; a set value overrides the default (Red Hat). |
| "Host patched, container safe." | No. The container carries its own PHP; glpi/glpi:latest still contains 8.5.12 on 21 November. |
"php -v shows whether I am patched." |
Only conditionally: the command line can have a different PHP version than FPM, and backport packages carry old version numbers with the fix. |
| "All 18 flaws are exploitable remotely without authentication." | No. CVE-2026-129291 is local, CVE-2026-129474 and CVE-2026-129182 affect only PHP 8.5, several need a configuration or a gadget chain. |
| "This is PHP2Shell again." | No. These are other CVEs; however, according to the advisories CVE-2026-129182 and CVE-2026-129481 are incomplete fixes for CVE-2026-109619. |
| "After the package update the new code is running." | Only after the restart: running processes keep the old code. |
Status and methodology
As of 21 November 2026, two days after publication. The information comes from the PHP ChangeLog, the release announcements and the GitHub advisories of php/php-src, the CVE records (for example CVE-2026-129289 and CVE-2026-115560), the Red Hat bulletin RHSB-2026-020 with the Security Data API, the trackers of Debian and Ubuntu, the CISA KEV catalog and the documentation of nginx and Apache. The default of session.upload_progress.enabled was read in the source code of the release tags.
The image data comes from the image configuration in the Docker registry (environment variable PHP_VERSION, creation date), the Dockerfiles and example configurations of the projects; for Snipe-IT from the package directory of the installed image layer, for the LinuxServer BookStack image from the project's package list. Images are rebuilt continuously; the dates are the creation dates from the image configuration, not necessarily the time from which a tag was available in the registry, and apply to the retrieval on 21 November 2026. Times of day for individual advisories, images and packages are not given; the date is what counts. "Not applicable" for CVE-2026-129289 describes a mod_php setup, "no keep-alive setting" describes the checked files of the project, not every installation. Not checked: whether applications open php://input without reading it fully, whether distribution packages restart the services themselves during the update (the Debian sources could not be retrieved because of bot protection), the function of the exploit repositories found on GitHub, and the documentation of moodle.org (bot protection). A BSI or CERT-EU advisory on the November flaws was not found in the search.
Our approach at WZ-IT
- Inventory: We record where PHP runs in your environment: PHP-FPM, mod_php, containers and appliances, plus the version the web server actually serves and how the front end is connected.
- Applying updates: We update system packages, pull base images again, rebuild your own images and restart PHP-FPM, Apache and containers.
- Reviewing configuration: For each pool and application we check
session.upload_progress,session.serialize_handler,error_logand the FastCGI settings of the front end. - Documenting: We record the running version per system and compare it with the advisory of the distribution or image vendor.
- Operations: In managed operations, updates, CVE monitoring and backups are included; the server audit checks existing systems once. Operated open-source applications are described under Managed Open Source and Managed Operations.
Further guides
- Why companies need CVE monitoring
- PHP flaws CVE-2026-109618 and CVE-2026-109619 (PHP2Shell): affected software and measures
- Java4Shell CVE-2026-114938: affected open-source applications
- OPNsense 26.1.8: two critical RCE bugs and patch management
- Nextcloud operations by WZ-IT
- Managed Open Source
- CVE monitoring
Check your PHP estate without finding out by chance: We record where PHP runs in your environment, apply the updates, restart the services and verify the running version. Book an initial consultation
Sources
- PHP: News Archive 2026 (announcement of 18 November, releases of 19 November, notice of 23 October)
- PHP: ChangeLog 8.x (8.5.13, 8.4.28, 8.3.37, 8.2.36)
- GitHub advisory GHSA-qc49-hpvq-g2w6: PHP-FPM request smuggling through an unread HTTP request body (CVE-2026-129289)
- GitHub advisory GHSA-658j-r8rw-q6mx: Session serializer mismatch (CVE-2026-115560)
- GitHub advisory GHSA-fm2m-fjf5-v3cp: Partially overwritten session files (CVE-2026-126714)
- GitHub advisory GHSA-g3vh-838c-65v7: Use-after-free after interrupted session shutdown (CVE-2026-129481)
- GitHub advisory GHSA-gr4f-wgx5-jpfv: Timezone cache use-after-free (CVE-2026-129496)
- GitHub advisory GHSA-8mjj-2r7p-4hg2: Use-after-free in magic method backtraces (CVE-2026-129474)
- GitHub advisory GHSA-2fjv-p558-g82m: last_fatal_error_backtrace (CVE-2026-129182)
- GitHub advisory GHSA-jm4m-qfqx-967m: PHP-FPM epoll use-after-free (CVE-2026-129291)
- CVE record CVE-2026-129289
- CVE record CVE-2026-115560
- Red Hat: RHSB-2026-020 Multiple vulnerabilities - PHP
- Red Hat: CVE-2026-129289
- Red Hat: Security Data API for CVE-2026-129289
- Red Hat: RHSA-2026:100821
- Debian Security Tracker: CVE-2026-129289
- Ubuntu Security: CVE-2026-129289
- CISA: Known Exploited Vulnerabilities Catalog
- nginx: fastcgi_keep_conn
- Apache: mod_proxy_fcgi
- PHP: Session configuration (manual)
- PHP source: main/SAPI.h (pre_request_init, 8.5.13)
- docker-library/php: Dockerfile 8.5 apache
- Nextcloud Docker: commit 7e9e152f
- Nextcloud All-in-One: Containers
- Nextcloud snap
- GLPI: docker-images













