Futureweb

Security

PHP's cleanup trusted request state that was already freed

PHP's October 21 archive names four releases for two critical flaws: cleanup trusted half-initialized request data, and a worker reused state after its memory was freed.

The short version

PHP says 8.5.12, 8.4.27, 8.3.36, and 8.2.35 fix two critical flaws that can allow unauthenticated remote code execution. One leaves request data half initialized when memory runs out. The other reuses request state after its memory was freed.

The archive replaced the schedule

This morning's brief recorded a schedule. PHP's October 20 notice had set a security release for October 21 at 14:00 UTC and had withheld the fixed version numbers. The archive entry dated October 21 names them. PHP 8.5.12, 8.4.27, 8.3.36, and 8.2.35 fix two critical vulnerabilities, CVE-2026-109618 and CVE-2026-109619. PHP says both can allow unauthenticated remote code execution, and it tells every PHP user to update immediately. SecurityWeek reports that several companies saw exploitation attempts before those packages were public. PHP says the update is the complete fix. Notes that only reduce exposure stay on the archive page.PHP told users to prepare for a critical security release.

Who the project says is exposed

PHP lists installations that serve web requests: PHP-FPM, Apache with mod_php, php-cgi, FrankenPHP, and LiteSpeed. It says the risk is highest for mod_php, because an unauthenticated request can reach the flaw before any application code runs. PHP 8.1 and older no longer receive updates from the project. The archive says some distributors and long-term support vendors ship backports, and that the security team posted minimal source backports for PHP 7.4, 8.0, and 8.1 for people who build from source. The package a vendor actually published is the copy an operator can install. A version guessed from a headline before the archive existed is not the list.The advisory is the story.

Lifetime, in the project's words

PHP describes CVE-2026-109618 as remote code execution through out-of-memory corruption. Out-of-memory handling can interrupt the interpreter while internal request data is only partly initialized. Request cleanup may then treat that invalid data as valid. PHP describes CVE-2026-109619 as remote code execution through stale global state. Request-owned internal state can survive after its backing memory has been freed. A later request on the same worker may use that stale state before normal initialization resets it. One bug uses data that was never finished. The other uses state after the memory was freed. The archive is the record of that class.What a CVE identifier is.

Safe Rust rejects this class at compile time

Safe Rust does not leave a half-built request object for a cleanup path to trust, and it does not let a later call read a worker global whose memory has already been freed. The borrow checker turns that program into a compile error. Remade with Rust rebuilds widely used primitives on that rule. The hot path is Rust. A wrapper that still calls the C interpreter leaves the flaw in the interpreter. The learn note on the rebuilds is the mechanism: bounds-checked slices, a borrow checker, and a use-after-free that does not compile in safe code. That is a property of the program you ship. It is not a patch script for a PHP tree you did not rebuild.Why Remade with Rust rebuilds the primitive.

The PHP host you still run

A service written in Rust does not update a PHP site that is still on the public internet. If the page the world hits is mod_php or PHP-FPM, the archive's four releases, or the distributor backport for an older branch, are the fix. An unsafe block can put a raw pointer back into freed memory. The house rule is that the block stays small and the reason is written down beside it. A codec, an allocator, or a parser rebuilt in safe Rust drops this class in that program. The neighbor process still running PHP is a separate host and a separate update.What an unsafe block has to name.

The same class, counted in Chrome

SecurityWeek's October 21 headline says memory bugs account for most of Chrome's 353 new security fixes. That figure is Chrome's, not a rerun of the PHP flaws, and it is not a quality table for this desk. It is the same class in a second product on the same day: lifetime errors in a large C++ codebase, closed one by one in a release. A CVSS number does not name the line that failed. PHP's archive does. Initialization was partial, or the worker kept state whose memory had already been freed. The 9.2 rating on the October 20 notice was the score. The archive is the mechanism.What a CVSS score does not say.

A MATA call has no PHP worker

MATA's public account is that the software is written in memory-safe languages and that keys stay on the holder's device, with encryption there, rather than a requirement to park the data on an external server. A call on that network is not accepted by mod_php, PHP-FPM, or a CGI worker that keeps process globals between requests. There is no pool in which one caller's freed request state can be read by the next caller. The PHP flaws need a PHP process that parses the request. That process is not how a MATA service admits a caller. The digital-freedom note is the published statement of that choice.How MATA keeps keys on the device.

A later call does not inherit the worker

The second flaw is an interpreter global that outlives the request that owned it. A session, on the terms already filed here, is not a bag of server memory the next request may trust. On this network the caller proves a key the holder still has, and the verifier checks that proof before the call belongs to anyone. A later call does not inherit a previous caller's interpreter leftovers, because the two calls do not share a PHP worker. Updating PHP remains the remedy wherever PHP is what answers the request. The mesh boundary is the absence of that worker on the path.What a session cannot be.

Match the branch you actually run

The October 20 notice was a schedule. The October 21 archive is the release. Operators who still run PHP match the branch they run to 8.5.12, 8.4.27, 8.3.36, or 8.2.35, or to the backport their distributor names. Linux vendors received the patches in advance, so a distribution package may not wear the upstream version number on the same afternoon. PHP says the volunteer repository it names will not have the packages today, and it points source builds at the backports for the branches the project no longer maintains. The version an operator can install is the one the vendor's own channel shows. On October 23 PHP published a separate informational note, CVE-2026-115560, and a full advisory waits for the next regular release. Those four versions are not that note.The later PHP note.

Sources

The reports this brief is filing. Futureweb did not republish them.

  1. PHP news archive
  2. SecurityWeek
  3. NCSC

Questions

Did PHP publish the fixed versions on October 21, 2026?

Yes. The archive names PHP 8.5.12, 8.4.27, 8.3.36, and 8.2.35 as the releases that fix the two critical flaws. The October 20 notice had withheld that list until the packages existed.

What class of bug does PHP assign to CVE-2026-109618?

PHP calls it remote code execution through out-of-memory corruption. Handling can stop while request data is only partly initialized, and cleanup can then treat that data as valid.

What class of bug does PHP assign to CVE-2026-109619?

PHP calls it remote code execution through stale global state. Request-owned state can survive after its memory is freed, and a later request on the same worker can use it.

Does a Rust rewrite install those PHP packages?

No. A site that still serves PHP needs the archive's releases or the distributor's backport. Safe Rust removes this lifetime class in the program you compile, which is a different host from mod_php.

Does a MATA service parse the public web with PHP?

No. MATA's published model is memory-safe code and keys that stay on the holder's device. A call is not accepted by a PHP worker that keeps freed request state for the next caller.