CVE-2024-42479
📋 TL;DR
This vulnerability in llama.cpp allows attackers to write arbitrary data to any memory address due to an unsafe pointer in the rpc_tensor structure. This can lead to remote code execution, memory corruption, or denial of service. Anyone using vulnerable versions of llama.cpp for LLM inference is affected.
💻 Affected Systems
- llama.cpp
📦 What is this software?
Llama.cpp by Ggerganov
⚠️ Risk & Real-World Impact
Worst Case
Remote code execution with full system compromise, allowing attackers to execute arbitrary commands, steal data, or pivot to other systems.
Likely Case
Denial of service through application crashes or memory corruption leading to unstable LLM inference services.
If Mitigated
Limited impact if proper network segmentation and least privilege access controls are implemented.
🎯 Exploit Status
The vulnerability involves direct memory manipulation which is typically straightforward to exploit once understood.
🛠️ Fix & Mitigation
✅ Official Fix
Patch Version: Commit b3561 or later
Vendor Advisory: https://github.com/ggerganov/llama.cpp/security/advisories/GHSA-wcr5-566p-9cwj
Restart Required: Yes
Instructions:
1. Update llama.cpp to commit b3561 or later
2. Recompile the application
3. Restart all llama.cpp services
🔧 Temporary Workarounds
Network Isolation
allRestrict network access to llama.cpp services to trusted networks only
Use firewall rules to block external access to llama.cpp ports
Service Account Hardening
linuxRun llama.cpp with minimal privileges to limit potential damage
sudo useradd -r -s /bin/false llamauser
sudo chown -R llamauser:llamauser /path/to/llama
🧯 If You Can't Patch
- Implement strict network segmentation to isolate llama.cpp from untrusted networks
- Deploy application-level firewalls or WAFs to monitor and block suspicious requests
🔍 How to Verify
Check if Vulnerable:
Check if your llama.cpp version is older than commit b3561 by examining the git commit hash or version informationCheck Version:
git log --oneline -1Verify Fix Applied:
Verify the commit hash includes b3561 or later, and test that the rpc_tensor structure no longer contains the unsafe data pointer📡 Detection & Monitoring
Log Indicators:
- Unexpected application crashes
- Memory access violation errors
- Unusual network connections to llama.cpp ports
Network Indicators:
- Suspicious payloads sent to llama.cpp RPC endpoints
- Unusual traffic patterns to inference services
SIEM Query:
source="llama.cpp" AND (event_type="crash" OR event_type="memory_error")