CVE-2026-119052: llama.cpp b2877 before b11935 ggml-rpc Server DoS via Oversized recv_msg Length
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 SETTENSOR with a length above std::vector::maxsize(), triggering an uncaught std::lengtherror that aborts the process and denies service to all clients.
Event History
Frequently Asked Questions
Which deployments are exposed to this denial-of-service condition?
Deployments running the ggml-rpc server in llama.cpp versions from b2877 before b11935 are affected if an attacker can reach the RPC service over the network. Because exploitation requires no authentication, any network-accessible rpc-server should be treated as exposed.
What must an attacker do to crash the server?
The attacker must complete the HELLO handshake and then send an RPC command, such as SET_TENSOR, with a message length larger than std::vector::max_size(). This causes an uncaught std::length_error that aborts the rpc-server process.
What is the operational impact of successful exploitation?
Successful exploitation crashes the rpc-server and denies service to all clients using it. The provided impact information indicates availability impact only, with no stated confidentiality or integrity impact.
How can I determine whether an instance is affected?
Check whether the deployment runs a llama.cpp build from b2877 through builds before b11935 and has the ggml-rpc server enabled. An affected server may terminate with an uncaught std::length_error after receiving a malformed oversized RPC message.