Denial of Service in h2oai/h2o-3

Vulnerability

Overview

In h2oai/h2o-3 version 3.46.0.2, a denial-of-service (DoS) vulnerability exists in the parsing of GZIP-compressed data files. The root cause is improper handling of highly compressed data, leading to significant data amplification when a GZIP file is uploaded and parsed repeatedly [1][2]. This flaw is located within the parsing logic, specifically in the ParseDataset.java file, where the server does not limit the expansion ratio of compressed input [3].

Attack

Vector and Exploitation

An attacker can exploit this vulnerability without authentication by uploading a specially crafted, highly compressed GZIP file to the H2O-3 server. By repeatedly triggering the parsing of this file, the attacker causes the server to allocate excessive memory and spawn a large number of concurrent slow-running jobs [1][2]. These jobs consume system resources, eventually causing the server to become unresponsive to legitimate requests.

Impact

Successful exploitation results in a complete denial of service. The H2O-3 server becomes unresponsive due to memory exhaustion and the accumulation of many slow-running parser jobs, effectively halting all machine learning and data processing tasks on that instance [1][2]. No data confidentiality or integrity is compromised, but the availability of the service is severely degraded.

Mitigation

As of publication, a fix for CVE-2024-7765 has not been released in a public advisory from H2O.ai [1]. Users are advised to monitor the H2O-3 repository for patch updates and, in the interim, restrict network access to the H2O-3 server and avoid exposing the parsing interface to untrusted users or the internet.