On 21 October 2026 the PHP project shipped emergency releases for two critical engine bugs, together nicknamed PHP2Shell. Two days on, the picture is clearer: the probing is enormous, the confirmed damage so far is zero, and on 23 October PHP published a third advisory, CVE-2026-115560, that the 21 October releases do not fix.
Below: what has been measured, why the one-line mitigation covers less than it seems, and what the new advisory asks of you. Not patched yet? Start with our patch-day explainer.
The basics in two sentences
CVE-2026-109618 and CVE-2026-109619 are memory-handling bugs in the PHP engine itself, both rated CVSS v4.0 9.2, that can let an unauthenticated web request run code on the server, and every release before 8.5.12, 8.4.27, 8.3.36 and 8.2.35 is affected (the 21 October advisory). Versions, backports and patch steps are in that article.
What the evidence shows after 48 hours
Amazon's MadPot honeypots recorded a PHP2Shell attempt on 17 October, four days before disclosure, and saw public proof-of-concept exploits from about 02:00 UTC on 21 October, twelve hours before the 14:00 UTC release. Cloudflare switched its WAF rules on at 03:02 UTC, and Patchstack logged exploitation reports from hosting partners at 07:34 UTC.
Volumes then went vertical. Cloudflare chief executive Matthew Prince said on 22 October it saw thousands of attempts in the ten hours before the announcement, tens of thousands in the 45 minutes after it, hundreds of thousands within two hours, and millions by 22 October (via TNW).
Attempts are not breaches. Sophos reported one successful exploitation at 16:18 UTC on 22 October and corrected itself at 21:21 UTC: its telemetry showed scanning and a small number of failed attempts, with no evidence of compromise. The blocked payload was a likely Mirai variant, Potassium.
Hunt.io watched one operator run ten rounds against 123 servers and exploit none, per SecurityWeek; SOCRadar and Wordfence report no confirmed shells either. As of the morning of 23 October, CISA had not added either CVE to its Known Exploited Vulnerabilities list.
The patch gap is the real problem. Qualys data from 12,000 customers, cited by SecurityWeek, shows about 199,000 PHP instances patched within twelve hours, roughly 110 million (71 percent) on end-of-life PHP 5.4 to 8.1 with no vendor fix coming, and about 44 million on supported branches not yet updated.
Why disabling upload progress covers only one path
The most repeated advice this week is session.upload_progress.enabled=0. Do it, but know its limits. CVE-2026-109619 works because stale global state survives an error and the next request on the same worker consumes it before PHP resets. The GHSA advisory confirms two consumers: session upload progress, and INI diagnostics applied before request start, like php_value lines in Apache .htaccess under mod_php or FPM frontends passing PHP_VALUE. The setting closes the first door only.
It does nothing for CVE-2026-109618, a separate out-of-memory bug reached through multipart parsing; PHP says there is no complete configuration mitigation, only updating. The setting must also go in php.ini, an FPM pool or the server configuration, followed by a worker restart; ini_set() and .user.ini run too late.
CVE-2026-115560: the third flaw
On 23 October PHP posted an informational advisory for CVE-2026-115560, a further remote code execution issue in session upload progress. It is present in the fully patched 8.5.12, 8.4.27, 8.3.36 and 8.2.35, has no published CVSS score or version list, and will be fixed in the next regular bugfix release. No exploitation has been reported.
All three preconditions must hold: config sets session.serialize_handler to php_binary or php_serialize, the application switches it to php at runtime with ini_set() before session_start(), and upload progress is still enabled. PHP calls it an unlikely RCE vector. The interim mitigation is the same setting as before; if it is already in place, nothing more is needed.
What to do now
- Patch to 8.5.12, 8.4.27, 8.3.36 or 8.2.35 and restart every PHP worker, Apache and FPM alike.
- Set
session.upload_progress.enabled=0inphp.ini, the FPM pool or the server config, restart, and keep it until the CVE-2026-115560 fix ships. On PHP 8.5 that cannot update yet, also setfatal_error_backtraces=0. - On 7.4, 8.0 or 8.1, get a vendor backport or upgrade; the official Docker PHP images for those branches carry no fix.
- Search access logs for multipart POSTs containing
PHP_SESSION_UPLOAD_PROGRESSwhere the app does not use it, memory errors on line 0, Apache segfaults and web workers spawning shells (Datadog lists them). A successful exploit can return HTTP 200. - If you changed
session.upload_progress.name, the WAF rule does not match. Skip unofficial "fix" plugins; Patchstack warns they are unverified and cannot patch the engine.
session.upload_progress.enabled=0
; PHP 8.5 only, while unpatched:
fatal_error_backtraces=0How Scantinel helps
Scantinel (scantinel.ai) is an automated website security assessment platform that runs OSINT, network and OWASP Top 10 web vulnerability tests against ownership-verified assets and delivers reports a business owner can act on. For PHP2Shell, the scan reads the PHP version your server exposes in headers and banners and flags runtimes that are end-of-life or below the fixed releases, so a forgotten site among the unpatched 44 million shows up quickly.
It cannot see inside the server: whether upload progress is off, which serialize handler your app sets, or whether workers were restarted are facts only you or your host can confirm. Find the hosts still advertising an old version, then patch and check the configuration yourself.
Comments
No comments yet.