llama.cpp denial of service in ggml-rpc message handling
ggml-org llama.cpp
ggml-org llama.cpp
| Product | Affected | Fixed in |
|---|---|---|
| ggml-org llama.cpp |
| No fix yet |
llama.cpp b2877 before b11935 contains an uncaught exception vulnerability in the ggml-rpc server that allows unauthenticated remote attackers to crash the rpc-server by sending oversized message lengths. After the HELLO handshake, attackers can send a command such as SET_TENSOR with a length above std::vector::max_size(), triggering an uncaught std::length_error that aborts the process and denies service to all clients.
From the CVE record · VulnCheck
The ggml-rpc server in llama.cpp is vulnerable to an uncaught exception when it receives an oversized message length. An unauthenticated remote attacker can send one after the HELLO handshake, crashing the rpc-server and denying service to all clients. No user interaction is described.
Engineers running the ggml-rpc server in llama.cpp versions b2877 before b11935 should check their deployments.