CVE-2024-0116

4.9 MEDIUM

📋 TL;DR

NVIDIA Triton Inference Server has an out-of-bounds read vulnerability where users can release shared memory regions while they're in use. This could allow attackers to cause denial of service by crashing the server. Organizations using Triton Inference Server for AI/ML inference are affected.

💻 Affected Systems

Products:
  • NVIDIA Triton Inference Server
Versions: All versions prior to 24.01
Operating Systems: Linux, Windows
Default Config Vulnerable: ⚠️ Yes
Notes: Affects deployments using shared memory regions for inference

📦 What is this software?

⚠️ Risk & Real-World Impact

🔴

Worst Case

Complete denial of service causing Triton server crashes and disrupting AI inference services

🟠

Likely Case

Service disruption through server crashes requiring restart, potentially affecting inference workloads

🟢

If Mitigated

Limited impact with proper access controls and monitoring in place

🌐 Internet-Facing: MEDIUM - Requires user access but could affect externally accessible inference endpoints
🏢 Internal Only: MEDIUM - Internal users with access could disrupt shared inference services

🎯 Exploit Status

Public PoC: ✅ No
Weaponized: UNKNOWN
Unauthenticated Exploit: ✅ No
Complexity: MEDIUM

Requires user access to trigger shared memory operations

🛠️ Fix & Mitigation

✅ Official Fix

Patch Version: 24.01 or later

Vendor Advisory: https://nvidia.custhelp.com/app/answers/detail/a_id/5565

Restart Required: Yes

Instructions:

1. Download Triton Inference Server 24.01 or later from NVIDIA NGC. 2. Stop current Triton server. 3. Deploy updated version. 4. Restart Triton server with updated configuration.

🔧 Temporary Workarounds

Disable shared memory

all

Configure Triton to not use shared memory regions

Set '--allow-shared-memory=system' to 'false' in Triton startup configuration

🧯 If You Can't Patch

  • Implement strict access controls to limit who can interact with Triton inference endpoints
  • Monitor Triton server logs for abnormal termination and implement automated restart procedures

🔍 How to Verify

Check if Vulnerable:

Check Triton server version with: tritonserver --version

Check Version:

tritonserver --version

Verify Fix Applied:

Verify version is 24.01 or later and test shared memory operations

📡 Detection & Monitoring

Log Indicators:

  • Unexpected Triton server crashes
  • Segmentation fault errors in logs
  • Abnormal termination of inference processes

Network Indicators:

  • Sudden drop in inference request responses
  • Connection resets to Triton endpoints

SIEM Query:

source="triton" AND ("segmentation fault" OR "crash" OR "abnormal termination")

🔗 References

📤 Share & Export