All experiments were performed on loopback interfaces with disposable containers and official
PHP images. No production system was targeted. The request-level chains were exercised with a
public PoC harness in a single-worker FPM pool; the OOM/RCE chain was exercised against official
php:8.4.26-apache and php:8.4.27-apache images with a prefork worker pool.
ASLR was controlled with setarch -R only in the ASLR-off runs. A debug build with the same
configure flags was used for symbol introspection; the official image does not expose the
internal symbols needed for offset calibration.
The measurement harness is published in lab/ so the numbers can be
re-derived without redistributing an exploit kit. lab/ ships the configs, probe scripts,
measurement modules and unit suite; the operational modules (C2, exfiltration, WAF evasion,
payload forgers, anti-forensics, multi-target scanner, unified orchestrator) are withheld by
design. Reference chain modules degrade to no-ops when the forge package is absent.
A Nextcloud 33.0.x test deployment running PHP 8.4.21 under Apache mod_php prefork was
used to validate the exposed configuration and patch level. The RCE proof was obtained on the
official php:8.4.26-apache image, not by executing the exploit against the Nextcloud
deployment. This distinction matters: the bug is in PHP request startup, before application
code, so the application name is not the primitive; the deployment's SAPI, worker model, body
limits, session settings and network path determine exploitability.
This research was conducted with multiple AI models and specialized subagents under human direction. The workflow used:
- an orchestrator model for planning, evidence tracking and final review;
- subagents for literature review, public PoC review, static source comparison, lab orchestration and documentation checks;
- independent re-checks of high-impact conclusions with a different model where possible.
AI output was treated as a hypothesis, not as evidence. Every finding in this repository is backed by a reproducible command/output, a public source, or a source-code comparison. Claims that could not be reproduced are marked as hypotheses, limitations or "not tested". No AI is listed as an author; the human author is JoshRob.
- No finding without a command, request, log excerpt or source line.
- Lab-only results are labeled as lab results.
- Public PoC claims are separated from independent measurements.
- Versions, SAPI, relevant
php.inivalues and ASLR state are recorded with each result. - Third-party exploit code is not redistributed.
The PHP-level primitive is reproducible on official images under the stated configuration.
Exploitability is not universal: it depends on the SAPI (mod_php prefork vs PHP-FPM), worker
count, session.upload_progress, memory_limit/post_max_size, ASLR, allocator and kernel
layout, build flags, and any proxy/WAF in front of PHP. The public ASLR-bypass PoC reproduced
on one host and not on another. The A/C/E request-level chains additionally depend on
application conditions (LFI, per-directory serializer mismatch, local file write).
Do not read "PHP-level bug" as "every PHP deployment is exploitable with the same PoC". Read it as "the application code is not the root cause; the deployment conditions decide whether the primitive is reachable and reliable".