twimco.Let’s talk ↗︎
Menu
Field notes / Commerce

PHP shipped security fixes this morning. Check six settings on your live site, then patch the branch you run

Your store runs on PHP, and PHP shipped security fixes this morning. The worst bugs need settings most sites don't use. One small page on your live site shows which ones you have, and later proves the update took.

Editorial illustration of two colleagues at a studio workbench in daylight, one sliding a small replacement module into an open, running server unit while the other compares a printed page with a laptop, beside a wire basket of wrapped parcels and a desk calendar with one day circled.

Update now, but stay in your branch: PHP 8.5.13, 8.4.28, 8.3.37 or 8.2.36, released November 19, 2026, fix up to 17 advisories, up to seven rated critical. PHP says the critical ones need non-default or uncommon setups, so check what the PHP serving your site uses: session storage and upload progress, error backtraces, error logging, and FastCGI connection reuse. Leave PHP 8.6.0 for after Black Friday. Restart PHP-FPM, then confirm the live site reports the new version.

This one depends on your settings

PHP put out security releases this morning, a month after PHP2Shell, and the first question is different this time.

In October the answer to "am I exposed?" was basically "yes, if you run PHP." Today it's "it depends on how your PHP is set up." When PHP announced this release yesterday, it said the worst issues "require either non-default configuration or uncommon setups." Now the details are out, and they back that up.

What shipped on November 19, a little after 9 AM Eastern:

  • PHP 8.5.13, 8.4.28, 8.3.37 and 8.2.36, one for each supported branch. The changelog lists 17 security advisories for 8.5, 14 for 8.4, 13 for 8.3 and 12 for 8.2.
  • Seven of them are rated critical on PHP's own GitHub advisories. Two of those only affect 8.5. On 8.2 it's four.
  • PHP 8.6.0, a new version of the language, came out the same morning. More on why that's a separate decision below.

As of about 9:45 AM Eastern, none of today's CVEs were on CISA's list of vulnerabilities known to be exploited. October's two PHP2Shell bugs have been on it since October 26. If you never installed October's fix, today's release covers both, and that's the more urgent reason to move.

You should update either way. But the settings decide whether it's a "today, between other things" job or a "drop everything" one. And reading them takes ten minutes, because the PHP behind your website can tell you.

Ask the PHP that serves your site, not the one in your terminal

Here's the trap. Type php -v on your server and you get an answer, but it comes from the command-line PHP. On Debian and Ubuntu that's usually a separate setup with its own php.ini (under /etc/php/8.3/cli/ rather than /etc/php/8.3/fpm/). Laravel's php artisan about and WP-CLI run through it too. Your shoppers never touch that PHP. They hit PHP-FPM, or Apache's mod_php, with different settings and sometimes a different version.

So ask the web server's PHP directly. Make a tiny page with a random name nobody would guess, upload it next to your site's index.php, open it once in a browser, and copy what it says:

```php
<?php // check-7f3a9c.php: delete as soon as you've read it
header('Content-Type: text/plain');
echo PHP_VERSION, ' ', PHP_SAPI, PHP_ZTS ? ' threaded' : '', ' pid ', getmypid(), "\n";
foreach (['session.save_handler', 'session.serialize_handler',
'session.upload_progress.enabled', 'zend.exception_ignore_args',
'fatal_error_backtraces', 'error_log', 'cgi.fix_pathinfo'] as $k) {
echo $k, ' = ', var_export(ini_get($k), true), "\n";
}
```

It prints the version, how PHP is running (fpm-fcgi, apache2handler or cgi-fcgi), and the settings these advisories hinge on. No passwords, no paths beyond the error log. Still, delete it once you've written the answers down. You'll put it back once, after the update.

On WordPress there's a shortcut for the first line: Tools → Site Health → Info → Server shows "PHP version" and "PHP SAPI" as the website sees them. It doesn't show the session settings, so you still want the page for the rest.

The PHP in your terminal and the PHP your customers hit can be two different things. Ask the one that serves the site.

Read it one line at a time

Here's what each line tells you, matched to PHP's advisories.

session.save_handler = 'files'. That's PHP's default, and it's what GHSA-fm2m-fjf5-v3cp is about. If writing a session file gets cut off partway, the file can end up half new and half old. If an attacker can get data into the session, cut the write off at the right moment, and your code has the right kind of exploitable classes, that can end in running code. That's a lot of ifs. If sessions live in Redis or the database, this particular bug doesn't apply.

session.upload_progress.enabled. If it says '1', keep reading. Upload progress is one way an outsider can push data into a session. It's also the setup for GHSA-658j-r8rw-q6mx, where session data written in one format gets read back in another. The researchers ran code that way on all four branches, using classes from Guzzle and Monolog, two very common libraries. The mismatch needs session.serialize_handler set to something other than 'php' somewhere. That could be in your code (grep -rn serialize_handler --include=*.php . finds it) or in another app that shares the same session folder. Some hosts switched upload progress off in October, and today's patch turns it off by default.

zend.exception_ignore_args and fatal_error_backtraces (PHP 8.5 only). If the first says '0' and the second is on, you match the setup GHSA-8mjj-2r7p-4hg2 describes: a crafted request that triggers a fatal error at the wrong moment can run code, although the researchers' demo also needed known memory addresses. PHP's recommended production php.ini sets zend.exception_ignore_args to on, which blocks the path the researchers demonstrated. The advisory names Laravel facades and Magento's guest shipping estimate as code that reaches the vulnerable spot, but says a full attack on either hasn't been shown. On 8.2 to 8.4 the second setting prints false, because it doesn't exist there.

error_log. A file path matters for GHSA-gr4f-wgx5-jpfv, which affects every branch. The trigger also needs PHP to log a warning while it starts up, such as a bad date.timezone or an outdated session setting. Look in your PHP log for warnings around the last restart. PHP says sending errors to syslog avoids the demonstrated trigger.

PHP_SAPI = cgi-fcgi. That's the old php-cgi binary, and with cgi.fix_pathinfo at its default of 1 it's covered by GHSA-5j3f-64r7-f564. PHP-FPM and mod_php aren't affected.

One more check, outside PHP. GHSA-qc49-hpvq-g2w6 is the one we'd look at first if you run your own web server. With a front end that keeps its connections to PHP-FPM open between requests, one crafted request can smuggle in a second one and run the attacker's PHP code. No memory tricks or race needed. Run grep -rn fastcgi_keep_conn /etc/nginx/ and, on Apache, look for enablereuse=on. PHP says standard configurations that close the connection after each request aren't affected by the demonstrated attack. If yours is on, turning it off is the stopgap until the update is live.

Decision guide

Which line on the check page raises the stakes

ConsiderationLook closer if you seeLower stakes if you see
Session storagesession.save_handler = 'files' and code that stores user-supplied text in the session.Sessions in Redis or the database (not affected by the file-write bug).
Upload progresssession.upload_progress.enabled = '1', especially with any non-'php' serialize_handler in code or a shared session folder.'0' everywhere that can write to the same session store.
Backtraces (8.5)zend.exception_ignore_args = '0' with fatal_error_backtraces on.zend.exception_ignore_args = '1', as in php.ini-production.
Error logA file path, plus warnings logged when PHP starts.syslog, or a clean startup with no warnings.
How PHP runscgi-fcgi (php-cgi) with cgi.fix_pathinfo = '1'.fpm-fcgi or apache2handler (not affected by the CGI bug).
Web servernginx fastcgi_keep_conn on, or Apache enablereuse=on, in front of PHP-FPM.Connections closed after each request (the standard setup).
Decision aid drawn from PHP's November 19, 2026 security advisories on GitHub. Not client data; every affected version should still update.

Update inside your branch. Leave 8.6 for January

If you're on 8.4, the fix is 8.4.28. If you're on 8.3, it's 8.3.37. During a Black Friday freeze, that's the change you want: the same version of the language with security fixes added, not a new one.

PHP 8.6.0 also includes these fixes, and it's tempting to make one jump. Don't, this week. PHP calls 8.6 "a major update of the PHP language," and its release notes list a page of deprecations and behavior changes. Some of them touch the parts of a store that are hardest to test on a Thursday. Sessions get stricter defaults, including SameSite=Lax cookies, which can change what happens when a payment page sends a customer back to your site. Plugins and PHP extensions may need their own updates first. Read the migration guide in January and give it a proper staging pass.

Two things change even inside the patch release. Both come from the session fixes, and both are worth knowing before you deploy:

  • Upload progress is now off unless you turn it on. If a plugin shows a progress bar using PHP's own upload progress feature, it stops until you set it back on yourself. If you do, PHP now throws an error when code changes session settings partway through a request.
  • On 8.3 and newer, a PHP worker now exits if its cleanup fails partway, instead of serving the next customer with leftover state. You might see a few more worker restarts in the log. That's the fix working.

Where to get the update depends on where your PHP came from. As of about 9:55 AM Eastern, the official Docker images had been updated in their source repository at 9:08 AM but the new tags still weren't on Docker Hub. Ondřej Surý's packages on deb.sury.org already showed 8.5.13, 8.4.28, 8.3.37 and 8.2.36; his Launchpad PPA still showed 8.4.27. Ubuntu and Debian hadn't posted PHP notices for these CVEs yet. If you're on a managed WordPress host, ask them when it'll be live, and then check the page yourself anyway.

If you're on PHP 8.2, patch to 8.2.36 now and put an upgrade on the calendar. PHP's support page ends 8.2 security fixes on December 31, 2026. On 8.1 or older, nothing is coming from PHP at all.

Then load the same page again

Install the update, then restart everything that runs PHP: PHP-FPM (systemctl restart php8.3-fpm on Debian and Ubuntu, with your version in the name), Apache with a full restart if you use mod_php, and any queue workers. Containers need a rebuild with docker build --pull and a fresh rollout. Our PHP2Shell note has more on restarts and on distro packages that keep the old version number.

Then put the check page back and load it again. You want three things:

  1. The first line shows the new version: 8.5.13, 8.4.28, 8.3.37 or 8.2.36. If you use Ubuntu's or Red Hat's own packages, the number may stay the same because they patch in place. In that case, check the package version against your distro's notice.
  2. The pid is different from this morning's. Same pid means the worker wasn't restarted, and it's still running the old PHP.
  3. session.upload_progress.enabled now says '0'. If it still says '1', something sets it explicitly. Find it, because that setting overrides the new default.

Load it a few times. Each load can land on a different worker, and they should all agree. If you have several servers or pods, check each one.

Then delete the page and write one line in the freeze log: old version, new version, time, who checked, and which settings you changed. It's the same rule as this week's WordPress, NGINX and Ruby fixes: installed isn't the same as running.

How the work moves

One page, read twice

  1. Read

    Load the check page through the website, not php -v. Note the version, SAPI, pid and settings.

  2. Stopgap

    Turn off FastCGI connection reuse or upload progress if they're on and nothing needs them.

  3. Patch the branch

    8.5.13, 8.4.28, 8.3.37 or 8.2.36 from the same source your PHP came from. Not 8.6 this week.

  4. Restart

    PHP-FPM or Apache (full restart), workers, and fresh containers built with --pull.

  5. Read again

    New version, new pid, upload progress '0' on every worker and server. Then delete the page and log it.

Illustrative workflow for one PHP site during a code freeze; not a client screenshot or measured result.

If you'd rather someone else checked

None of this takes long. It's just one more runtime update in the week everyone wanted quiet, and the settings that matter are spread across php.ini, pool files, the web server and the app's own code.

If your store or app runs on WordPress, WooCommerce, Laravel or anything else built on PHP, we can do the check with you: read the settings, patch the branch you run, confirm every server is on it, and agree a date for the bigger upgrade once the holidays are over. That's the kind of production work our ecommerce team does, not only the storefront. Tell us what you're running.

Related work: Paddle8.

twimco.

We build alongside the people doing the work.
Bring us the problem you’re thinking about.

Related work

Paddle8 ↗︎