One month after PHP2Shell, PHP shipped another batch of security fixes. And this one is bigger: 17 vulnerabilities, 7 of them critical.
On November 19, PHP 8.5.13, 8.4.28, 8.3.37 and 8.2.36 came out, all of them security releases. PHP 8.6.0 came out the same day and already ships with all of these fixes.
If you don't have time to read the rest, take away just this: update to 8.5.13, 8.4.28, 8.3.37 or 8.2.36 (or to 8.6.0) and restart your workers. And if you run PHP-FPM with persistent connections or the official Docker image, read the FPM part carefully.
Contents
- What came out on November 19
- The critical flaws
- The high and medium flaws
- How does this relate to PHP2Shell?
- How do I know if I'm vulnerable?
- I can't update right now, what now?
- What about unsupported versions?
- Conclusion
What came out on November 19
PHP announced a day in advance that the release would affect all supported versions and that the highest severity was critical, but that the critical issues require non-default configuration or uncommon setups. That matters: unlike PHP2Shell, which hit practically any PHP serving HTTP, most of the current flaws depend on some specific condition.
According to the advisories published on the php-src GitHub, there are 7 critical, 4 high and 6 medium flaws (CVSS v4.0), plus the update to the uriparser library bundled in ext/uri, which the changelog lists as CVE-2026-48841. That's why some articles mention 18 flaws. Others mention 4 critical ones, but the official advisories rate 7 as critical.
As of this writing, there are no reports of active exploitation of these flaws, and none of them is in CISA's KEV catalog. Even so, there are already reports of public proofs of concept for some of them. And after PHP2Shell, I wouldn't count on that lasting long.
The critical flaws
All the critical ones were rated CVSS v4.0 9.2.
CVE-2026-129289: request smuggling in PHP-FPM
This is the most serious one in the batch, and the one that deserves the most attention from anyone running FPM.
When the application opens php://input but doesn't read the request body to the end, PHP stops discarding what's left. With a malformed multipart body, the attacker's bytes are left on the connection. If the web server reuses the connection to FPM (FCGI_KEEP_CONN), FPM parses those bytes as a new FastCGI request, with forged parameters such as SCRIPT_FILENAME, REMOTE_ADDR and PHP_ADMIN_VALUE.
From there, if the attacker knows the path of some PHP script that exists on the server, they just need to turn on allow_url_include and point auto_prepend_file at php://input to run PHP code:
It's unauthenticated RCE in a single request, with no memory corruption involved. And the changed settings can stick around on that worker.
The good news: it only affects setups that reuse connections to FPM, such as fastcgi_keep_conn on in Nginx, enablereuse=on in Apache, or HAProxy with persistent connections. The default Nginx and Apache configurations are not affected. The vulnerable code dates back to PHP 5.6. Details in advisory GHSA-qc49-hpvq-g2w6.
CVE-2026-115560: mismatched session serializers
This one was already announced on October 23, right after PHP2Shell, and now it's fixed. Session upload progress writes progress data with one serializer (php_binary or php_serialize), and it can later be read with another (php). A | in the value sent as PHP_SESSION_UPLOAD_PROGRESS, which PHP uses in the session key, becomes a delimiter and the content reaches unserialize(). The result is object injection, which becomes RCE if the application has a gadget chain available. The advisory mentions chains reproduced with Guzzle and Monolog.
It requires upload progress to be enabled and the serializer to change at runtime or differ between applications, pools or CLI jobs sharing the same session store.
The fix brings a behavior change: session.upload_progress.enabled is now disabled by default, and changing session settings at runtime while it's enabled is now rejected. If you use an upload progress bar, you'll need to enable it explicitly. Advisory: GHSA-658j-r8rw-q6mx.
CVE-2026-126714: phpAfterShell
This one was found by depthfirst while reviewing the PHP2Shell fix itself, which is why it got the nickname phpAfterShell.
The files session handler, PHP's default, didn't truncate the file before writing. An interrupted write left the start of the new session followed by the end of the old one, and that mix could deserialize attacker objects. The advisory says RCE depends on a gadget chain, but depthfirst claims it built the full chain using only PHP's native classes, without depending on the application. The remote path also goes through upload progress, so the same mitigation as the previous one applies here. Advisory: GHSA-fm2m-fjf5-v3cp.
CVE-2026-129496: use after free in the timezone cache
In NTS builds, a startup warning written to a file-based error_log, such as an invalid date.timezone, leaves a dangling pointer. If a fresh worker emits another warning early in a request, such as the max_input_vars one, it uses that pointer. Workers can be crashed remotely, and RCE was demonstrated locally, with known addresses and extra configuration. If your startup log has warnings, fix them anyway. Advisory: GHSA-gr4f-wgx5-jpfv.
The other three critical flaws are direct consequences of PHP2Shell, and I cover them further down.
The high and medium flaws
CVE-2026-129291: privilege escalation in PHP-FPM
Rated 8.8, but with a detail that catches a lot of people: with catch_workers_output enabled, code running as the pool user can execute commands as the FPM master process, which usually runs as root.
The default for catch_workers_output is no, but the official PHP Docker images turn it on, and the flaw was reproduced on them without any modification. It's a local attack, meaning the attacker already needs to run code in the pool. But combined with any other flaw in the application, it becomes the path to taking over the whole container. Advisory: GHSA-jm4m-qfqx-967m.
CVE-2026-128064: uppercase HTTPS:// goes out unencrypted
This one is curious. PHP's HTTP and FTP stream wrappers didn't recognize the scheme when it was written in uppercase, so the request was sent in plain text:
// Before the fix, this went out without TLS
$response = file_get_contents('HTTPS://api.example.com/orders?token=secret');
It also applies to FTPS:// and to redirect targets, which means a server can redirect your request to a URL with an uppercase scheme and receive everything unencrypted. Tokens, passwords and the request body are exposed to anyone on the path. Rated 8.2. Advisory: GHSA-f6gg-rchx-xw8j.
The rest
| CVE | Component | What it is | Score |
|---|---|---|---|
| CVE-2026-127960 | HTTP wrapper | "Folded" headers let a malicious server pin the CPU (DoS) | 8.7 |
| CVE-2026-128090 | Phar | Switching the signature algorithm bypasses the OpenSSL key | 7.7 |
| CVE-2026-129538 | php-cgi | Out-of-bounds read and write when fixing up PATH_INFO |
6.9 |
| CVE-2026-15353 | SOAP | 4-byte heap leak | 6.3 |
| CVE-2026-129506 | SOAP | Use after free in DTD entities (DoS) | 6.3 |
| CVE-2026-129501 | DOM | Double free in saveXml() on Dom\XMLDocument and Dom\HTMLDocument (8.4 and 8.5) |
6.3 |
| CVE-2026-129438 | URI | Overflow when serializing ~2 GiB URIs (8.5 only) | 6.3 |
| CVE-2026-129292 | PHP-FPM | XSS on the status page | 5.3 |
The php-cgi one doesn't affect FPM or mod_php. The DOM one only affects the new API (Dom\XMLDocument and Dom\HTMLDocument), not the old DOMDocument.
How does this relate to PHP2Shell?
Quite a lot. Of the 7 critical flaws, four came directly out of the October case, and a fifth was announced alongside it:
The first three are memory bugs (use after free) in shutdown and backtraces. The two 8.5-only ones depend on fatal_error_backtraces, which is the 8.5 default. CVE-2026-129474 also requires zend.exception_ignore_args=0, and php.ini-production already ships with 1, which blocks the attack. Details in advisories GHSA-g3vh-838c-65v7, GHSA-2fjv-p558-g82m and GHSA-8mjj-2r7p-4hg2.
That doesn't mean the October fix was poorly done. It's what usually happens after a big flaw: everyone, researchers and attackers alike, goes over the same code with a magnifying glass, often with help from AI. The PHP2Shell fix touched more than 300 files, so it was to be expected that some corners would still be uncovered.
How do I know if I'm vulnerable?
You're protected if you're on one of these versions or later:
| Branch | Fixed version |
|---|---|
| 8.6 | 8.6.0 |
| 8.5 | 8.5.13 |
| 8.4 | 8.4.28 |
| 8.3 | 8.3.37 |
| 8.2 | 8.2.36 |
| 8.1 or earlier | No fix from PHP |
Here's the same check from the PHP2Shell post, updated. Create a temporary file on the server, open it in the browser (so it runs under the same SAPI as the site) and delete it right after:
<?php
$fixed = ['8.6' => '8.6.0', '8.5' => '8.5.13', '8.4' => '8.4.28', '8.3' => '8.3.37', '8.2' => '8.2.36'];
$branch = PHP_MAJOR_VERSION . '.' . PHP_MINOR_VERSION;
echo 'Version: ', PHP_VERSION, ' | SAPI: ', PHP_SAPI, PHP_EOL;
if (!isset($fixed[$branch])) {
echo 'Unsupported branch: there is no official release with the fix', PHP_EOL;
} elseif (version_compare(PHP_VERSION, $fixed[$branch], '<')) {
echo "VULNERABLE: update to {$fixed[$branch]} or later", PHP_EOL;
} else {
echo 'Version includes the fix', PHP_EOL;
}
// Settings that increase exposure until you update
echo 'session.upload_progress.enabled: ', ini_get('session.upload_progress.enabled') ?: '0', PHP_EOL;
echo 'zend.exception_ignore_args: ', ini_get('zend.exception_ignore_args') ?: '0', PHP_EOL;
If you use PHP-FPM, also check the web server and the pool:
# Nginx: persistent connections to FPM (CVE-2026-129289)
grep -rn "fastcgi_keep_conn" /etc/nginx/
# FPM pool: catch_workers_output (CVE-2026-129291)
grep -rn "catch_workers_output" /usr/local/etc/php-fpm.d/ /etc/php/*/fpm/pool.d/ 2>/dev/null
The same precautions from October still apply:
- distro packages (Debian, Ubuntu, RHEL) backport fixes without changing the version number. Check your distribution's security advisory. Red Hat, for example, rated CVE-2026-129289 as critical;
- Docker images don't update themselves: run
docker pull, rebuild and deploy; - restart all workers after updating or changing any setting. Old processes keep running the old binary and the old
php.ini.
I can't update right now, what now?
These mitigations cover the most serious flaws. They reduce the risk, but they don't replace updating.
In php.ini:
session.upload_progress.enabled = 0 ; CVE-2026-115560 and CVE-2026-126714
zend.exception_ignore_args = 1 ; CVE-2026-129474 (PHP 8.5)
fatal_error_backtraces = 0 ; CVE-2026-129182 (PHP 8.5)
In the FPM pool (www.conf or equivalent):
catch_workers_output = no ; CVE-2026-129291
php_admin_flag[session.upload_progress.enabled] = off ; if the pool overrides php.ini
Set these options in php.ini or with php_admin_*, never with ini_set() in the application: by then the request has already been processed and it's too late.
On the web server, turn off connection reuse to FPM (fastcgi_keep_conn off in Nginx, no enablereuse=on in Apache). That closes CVE-2026-129289, the most dangerous one on the list. For CVE-2026-129496, fix the startup warnings or send error_log to syslog.
After changing any of these options, restart all workers.
What about unsupported versions?
In October, PHP itself published backport branches for 7.4, 8.0 and 8.1. This time, PHP published nothing for unsupported versions. And the advisories list PHP 8.1 as affected with no fix for at least two flaws: the FPM privilege escalation and the HTTP wrapper DoS. Since several of the flaws have existed since PHP 4 or 5, older versions are very likely affected by almost all of them.
The good news comes, once again, from Ondřej Surý, who maintains the ppa:ondrej/php and deb.sury.org packages. He has already backported the fixes to PHP 5.6 through 8.1 as new package revisions. The version number doesn't change (php -v still shows 8.1.34, for example), so what matters is the package revision:
| Version | Fixed revision (or later) |
|---|---|
| 8.1 | 8.1.34-13 |
| 8.0 | 1:8.0.30-30 |
| 7.4 | 1:7.4.33-36 |
| 5.6 to 7.3 | Also got new revisions |
To check:
apt update && apt upgrade
dpkg -l 'php8.1*' | awk '/^ii/ {print $2, $3}'
If you use a mirror of the PPA, the packages may take a few hours to arrive. Ubuntu Pro has also published fixes for the old packages, from PHP 5 to 7.4, but not yet for every CVE.
That solves the immediate problem, but it's still a stopgap: it depends on one person's volunteer work and doesn't replace being on a supported version.
And a reminder: PHP 8.2 reaches end of life on December 31, 2026, a little over a month from now. The next batch of fixes affecting 8.2 in 2027 won't get a release. If you're still on it, now is the time to plan the migration. And PHP 8.6 just came out with plenty of good stuff.
Conclusion
This release didn't make as much noise as PHP2Shell, but it has flaws just as serious, only more dependent on configuration. The FPM one with persistent connections is what I'd look at first, followed by the privilege escalation in the official Docker images.
The pace also stands out: two big batches of critical CVEs in one month. That has a lot to do with the current moment, in which AI has sped up the discovery of old flaws across the whole ecosystem, not just in PHP. The good side is that PHP is responding within its regular release cycle, with detailed advisories and clear mitigations.
The lesson is the same as in October: update, restart your workers and keep your environment in a state where updating is routine, not an emergency.
Until next time, cheers! 🐘
References:
- PHP.net: release announcements for 8.5.13, 8.4.28, 8.3.37, 8.2.36 and 8.6.0
- PHP.net: PHP 8 ChangeLog
- GitHub: php-src security advisories
- PHP.net: supported versions
- Precursor Security: phpAfterShell, critical PHP RCE vulnerabilities
- Techzine: PHP security updates fix 18 vulnerabilities
- Chainguard: PHP's November security release, what it is and what to update
- Red Hat: CVE-2026-129289