CVE-2025-71006: OneFlow v0.9.0 DoS Vulnerability

CVE-2025-71006 Overview

A floating point exception (FPE) vulnerability exists in the oneflow.reshape component of OneFlow v0.9.0. This vulnerability allows attackers to cause a Denial of Service (DoS) condition by supplying specially crafted input to the affected reshape function. The vulnerability is classified under CWE-369 (Divide By Zero), indicating the root cause involves improper handling of division operations that can result in floating point exceptions.

Critical Impact

Attackers can remotely trigger application crashes and service disruption in systems using OneFlow v0.9.0 for machine learning operations by providing malicious input to the reshape function.

Affected Products

• OneFlow v0.9.0
• Systems utilizing the oneflow.reshape component
• Machine learning pipelines and applications built on affected OneFlow versions

Discovery Timeline

• 2026-01-28 - CVE-2025-71006 published to NVD
• 2026-01-29 - Last updated in NVD database

Technical Details for CVE-2025-71006

Vulnerability Analysis

This vulnerability stems from improper handling of floating point operations within the oneflow.reshape component. When processing certain input parameters, the reshape function fails to properly validate divisor values before performing calculations, leading to a floating point exception (FPE). This type of exception occurs when the processor encounters an invalid arithmetic operation, such as division by zero or other undefined floating point calculations.

The network-accessible nature of this vulnerability means that remote attackers can exploit it without requiring any privileges on the target system, though user interaction is required to trigger the vulnerable code path. Successful exploitation results in complete disruption of the availability of the affected application.

Root Cause

The vulnerability is rooted in CWE-369 (Divide By Zero), indicating that the oneflow.reshape component does not adequately validate input parameters before using them in division operations. When a zero or invalid value is passed as a dimension parameter during tensor reshape operations, the function attempts to perform calculations that result in a floating point exception, causing the application to crash.

Attack Vector

The attack is conducted over the network and requires user interaction to execute. An attacker can craft malicious input data designed to trigger the reshape operation with parameters that cause a division by zero scenario. When a user or automated system processes this malicious input through OneFlow's reshape functionality, the floating point exception is triggered, resulting in immediate application termination.

The vulnerability mechanism involves providing specially crafted tensor dimensions or reshape parameters that, when processed by the oneflow.reshape function, cause the underlying arithmetic operations to encounter an FPE condition. Technical details regarding the specific exploitation method can be found in GitHub Issue #10653.

Detection Methods for CVE-2025-71006

Indicators of Compromise

• Unexpected application crashes or service terminations in OneFlow-based applications
• SIGFPE (floating point exception) signals in system logs associated with OneFlow processes
• Abnormal or malformed tensor reshape requests in application logs
• Repeated crash-restart cycles in machine learning pipeline services

Detection Strategies

• Monitor application logs for floating point exception errors originating from OneFlow components
• Implement crash dump analysis to identify FPE-related terminations in the oneflow.reshape function
• Deploy anomaly detection for unusual input patterns targeting tensor reshape operations
• Configure application-level monitoring to alert on repeated crashes in ML pipeline components

Monitoring Recommendations

• Enable detailed logging for OneFlow reshape operations to capture input parameters
• Set up automated alerts for SIGFPE signals in containerized or standalone OneFlow deployments
• Monitor resource utilization patterns that may indicate repeated crash-restart cycles
• Implement input validation logging at application boundaries before data reaches OneFlow components

How to Mitigate CVE-2025-71006

Immediate Actions Required

• Audit applications using OneFlow v0.9.0 to identify usage of the oneflow.reshape component
• Implement input validation to sanitize tensor dimensions before passing to reshape operations
• Consider isolating OneFlow workloads to limit the impact of potential DoS attacks
• Monitor the official OneFlow repository for security patches and updates

Patch Information

At the time of publication, users should monitor the official OneFlow GitHub repository for security updates. The vulnerability has been documented in GitHub Issue #10653, which may contain information about fixes or patches as they become available.

Workarounds

• Implement application-level input validation to reject zero or invalid dimension values before reshape operations
• Add try-catch exception handling around reshape function calls to gracefully handle FPE conditions
• Deploy rate limiting on endpoints that accept tensor reshape parameters to mitigate DoS impact
• Consider using alternative tensor manipulation methods that include built-in boundary checking

# Example input validation before reshape operations
# Validate tensor dimensions are non-zero before calling reshape
if [ "$TENSOR_DIM" -eq 0 ]; then
    echo "Error: Invalid tensor dimension - cannot be zero"
    exit 1
fi