Skip to main content
CybersecurityVulnerability Management

Researcher Exploits Database Flaws in PostgreSQL, MySQL, Oracle

Researcher working at a lab bench with computer equipment and code on a monitor.

"A database account's SQL permissions control which tables and schemas it can access, the researcher argues, not the memory-unsafe C and C++ code the engine runs to read, plan, and execute a query," researcher Zheng Yu wrote in published proofs of concept this week.

That distinction is the throughline of a compact but consequential piece of research: a low‑privilege database account — one that cannot read an application's tables — can nonetheless reach memory‑corruption bugs in the database engine itself and turn those bugs into code running on the database server. Zheng Yu published proofs of concept for PostgreSQL, MySQL, ClickHouse, and a separate Oracle chain; none of the exploit chains has been seen in a real attack, and the full chains were not reproduced independently.

Zheng Yu's demonstrations and the underlying technique

The researcher showed how common grants and configuration options give an otherwise constrained account the inputs it needs to trigger memory‑safety bugs in engine code written in C and C++. In PostgreSQL, an account with the ability to create database objects can build the objects it needs in a temporary schema and then convert a type‑confusion bug into control of the server. In MySQL and ClickHouse the ability to create temporary tables or to change a single format option provided the pathway into engine code. Oracle required a different path: chained flaws in Oracle Text and the GraphQL interface of Oracle REST Data Services reached engine code via an exposed web endpoint, and — according to the researcher — required no login at all.

PostgreSQL and MySQL: the vulnerabilities and vendor responses

PostgreSQL’s issue was assigned CVE‑2026‑116331. The researcher’s chain allows an account that can create database objects to run code as the operating system user that runs the database; a fix was issued last week in a security advisory, and the CVE record credits Zheng Yu for the report.

MySQL’s problem is in the query optimizer and has been tracked as CVE‑2026‑101786 in Server and Cluster versions 26.7.0 to 26.7.1. Oracle’s October advisory rated that flaw as requiring high privileges, but the researcher reached code execution on a 26.7.0 image using an account that could only create temporary tables and could not read existing ones — a narrower access profile than the vendor’s advisory suggests.

Oracle and ClickHouse: no-login chain and a still-unfixed bug

Oracle’s chain combines an Oracle Text bug, CVE‑2026‑114819, and a GraphQL flaw in Oracle REST Data Services, CVE‑2026‑91609. Oracle patched the Oracle Text bug in its November update and patched the GraphQL flaw in October. The company, on its own advisories, rated neither bug as exploitable without a login; the researcher’s chain, however, starts from an exposed web endpoint and — according to the published work — reaches engine code before any login.

ClickHouse presented an integer‑overflow write reachable by a user who sets the option format_binary_max_string_size to zero; Zheng Yu reproduced a shell on version 26.7.5.10. That ClickHouse issue has no CVE and no confirmed fixed version; the researcher and advisories recommend operators check with the vendor for a fix.

Current status: patches, confirmation, and real‑world use

Four of the flaws now have vendor CVEs and fixes, and Oracle has issued October and November updates for the two components it affected. But the published chains were not reproduced independently, and the researcher reports no sign that any of them has been used in a real attack. ClickHouse remains an exception: without a confirmed CVE or fixed version, operators have a remaining action item to verify vendor guidance.

What this means for technologists, operators, and enterprise leaders

  • Technologists and security teams: Install the available updates for PostgreSQL, MySQL, and Oracle components; verify whether your installed versions match those the advisories addressed.
  • Database administrators and operators: Review not only what accounts can read but what they can create or change — in particular, remove the TEMPORARY privilege and broad temporary‑table rights from accounts that do not need them.
  • Procurement and IT leadership at affected enterprises: Confirm fixed versions with vendors for ClickHouse and ensure patches for the disclosed CVEs have been applied in deployed instances.

Practical steps are clear and narrow: apply vendor fixes where available, check versions, and tighten the create‑object and temporary‑table rights on accounts that don't require them. That last measure will not close the Oracle chain that begins without any account, however, which remains the clearest reason for teams to check both network exposure and vendor advisories.

The researcher's work is a reminder that SQL permissions and memory‑unsafe engine code are different layers, and that ordinary grants such as TEMPORARY can bridge them. Operators should treat the patched CVEs as urgent and the ClickHouse path as unresolved until vendors confirm a fix.

Original story: https://thehackernews.com/2026/11/researcher-demonstrates-database-rce.html