ambionics Exploits and Proofs of Concept

Showing 5 PoCs on this page

GitHub

ambionics/cve-2026-9082-drupal-postgresql-rce

Repository PoC11Created 2026-05-26ExploitCVE-2026-90828 files

18.7 KiB

Analysisdeepseek-v4-pro:cloud · 2026-08-07

Technical assessment

A Python exploit that leverages an unauthenticated SQL injection in Drupal's JSON:API to achieve remote code execution on a PostgreSQL backend by writing a native preload library via large objects and triggering it through session_preload_libraries.

Backdoor review

No backdoor observed in reviewed code

The repository contains a proof-of-concept exploit for CVE-2026-9082, a Drupal PostgreSQL SQL injection. The Python script performs the documented exploit steps: SQL injection, PostgreSQL preload library compilation, file write via large objects, configuration reload, and command output retrieval. No concealed backdoor, unrelated payload, or deceptive behavior was observed in the reviewed text files.

ClassificationExploit

Model confidence98%

AuthenticationNot required

LanguagesPythonCShell

Target softwareDrupal corePostgreSQL

Attack typesSQL InjectionRemote Code Execution

Evidence & reasoningClassification basis · observed behavior · safety review

Technical evidence

Classification basis and observed behavior

Classification basis

The artifact contains a complete, functional exploit that actively exercises the SQL injection vulnerability to achieve remote code execution. It sends malicious SQL payloads, writes files to the server, and executes arbitrary commands, which is the definition of exploit code.

README.md:3drupal-postgres-preload-rce.py:1-356

Requirements

Target Drupal instance with vulnerable JSON:API endpoint and PostgreSQL superuser connectionREADME.md:33
Local C compiler (cc) to build a PostgreSQL preload shared objectREADME.md:40
Python 3 with requests libraryrequirements.txt:1

Observed behavior

Sends crafted HTTP GET requests to /jsonapi/node/article with SQL injection payloads in filter parametersdrupal-postgres-preload-rce.py:162-171
Extracts PostgreSQL version, current user, superuser status, and data directory via SQL injectiondrupal-postgres-preload-rce.py:303-306
Compiles a C shared library containing a _PG_init function that executes an arbitrary shell commanddrupal-postgres-preload-rce.py:212-246
Writes the compiled shared library to the PostgreSQL data directory using lo_create, lo_put, and lo_exportdrupal-postgres-preload-rce.py:256-264
Modifies postgresql.auto.conf to set session_preload_libraries and dynamic_library_path, then reloads configurationdrupal-postgres-preload-rce.py:267-279drupal-postgres-preload-rce.py:330-331
Triggers a new PostgreSQL backend to load the preload library and execute the command, then reads output via pg_read_filedrupal-postgres-preload-rce.py:334-345
Restores original postgresql.auto.conf settings after exploitationdrupal-postgres-preload-rce.py:348-352

Safety-review evidence

Behaviors behind the backdoor verdict

Observables

Exploit Behavior: Payload withheldThis is the expected behavior of a PoC for the described CVE. The command is provided by the user at runtime and is not hardcoded.drupal-postgres-preload-rce.py:294-356
Command Execution: Payload withheldThis is the RCE mechanism of the exploit. The command is not fixed; it is passed as a command-line argument.drupal-postgres-preload-rce.py:212-226
Cleanup Behavior: Payload withheldThis indicates an attempt to clean up after exploitation, not to establish persistence.drupal-postgres-preload-rce.py:348-352

Review boundaries

What the analysis did not establish

Only three of eight repository files were provided as text; five files (docker-compose.yml, Dockerfile, entrypoint.sh, seed-content.php, and one other) are omitted, so the full lab setup and auxiliary scripts are not visible.
The analysis is based solely on static source code review; the exploit was not executed, and its runtime behavior is inferred from the code and documentation.
5 unclassified files (binary or non-text) were present in the repository but not analyzed; their content is unknown.
The review is limited to the supplied text evidence and does not include dynamic analysis or execution of the code.

Model interpretation

This review is limited to the supplied PoC code and context. It does not assert that the code works or is safe to execute.

GitHub