Executive Summary

This vulnerability exists in the sgl-project SGLang up to version 0.5.11, specifically in the data_hash function of the Cache Handler component.

It allows an attacker to cause a denial of service (DoS) condition by manipulating this function.

The attack requires local access to the system and a high degree of complexity, making exploitation difficult.

Although the exploit has been publicly disclosed, a fix is pending acceptance via a pull request.

Useful 0 Not Useful 0

Impact Analysis

The primary impact of this vulnerability is the potential to cause a denial of service (DoS) on the affected system.

This means that an attacker with local access could disrupt the normal operation of the Cache Handler component, potentially affecting system availability.

Due to the high complexity and local access requirement, the risk of exploitation is limited but still present.

Useful 0 Not Useful 0

Mitigation Strategies

The vulnerability affects the function data_hash of the Cache Handler component in sgl-project SGLang up to version 0.5.11 and allows denial of service via local execution.

Since the exploit requires local access and a high degree of complexity, immediate mitigation steps include restricting local access to trusted users and environments.

Additionally, monitor for any updates or acceptance of the pull request that fixes this issue and apply the patch as soon as it becomes available.

Useful 0 Not Useful 0

Compliance Impact

The provided information does not specify any direct impact of this vulnerability on compliance with common standards and regulations such as GDPR or HIPAA.

Useful 0 Not Useful 0

Detection Guidance

This vulnerability is related to local execution in the sgl-project SGLang's Cache Handler component, specifically the data_hash function causing denial of service through hash collisions. Detection involves identifying abnormal crashes or denial-of-service symptoms in local SGLang deployments, especially related to multimodal embedding cache corruption.

Since the attack is local and involves hash collisions causing embedding corruption and crashes, detection can focus on monitoring logs for cache-related errors, crashes, or TypeErrors in the hash_feature function when handling non-tensor lists.

No specific network commands are applicable because the attack vector is local execution with high complexity.

Suggested commands for detection might include:

• Checking application logs for error messages related to cache or embedding corruption.
• Using system monitoring tools to detect unexpected crashes or high resource usage in SGLang processes.
• Running diagnostic scripts or tests that simulate tensor hashing to observe if hash collisions or TypeErrors occur, based on the minimal script demonstrated in Resource 3.

However, no explicit detection commands or scripts are provided in the available resources.

Useful 0 Not Useful 0