CVE-2026-73511
Description détaillée
Envoy is an open source edge and service proxy designed for cloud-native applications. Prior to 1.36.10, 1.37.6, 1.38.4, and 1.39.1, Envoy normally matches the raw request path, while servlet backends such as Apache Tomcat strip semicolon matrix parameters from each path segment before resolving the resource. Envoy's ignore_path_parameters_in_path_matching option instead truncates at the first semicolon and still does not match per-segment backend behavior. A remote client can use a parameterized protected segment, or a parameter on an earlier segment, to make Envoy select an unprotected fallback while the backend resolves the protected resource. The relevant scope boundary is that the bypass requires both a path-based Envoy decision and a backend that strips semicolon parameters per segment. This issue is fixed in versions 1.36.10, 1.37.6, 1.38.4, and 1.39.1.
Vecteur d'attaque (CVSS)
Références et Patchs
Dernières Vulnérabilités
CVE-2026-94627
vLLM Mooncake connector through 0.29.0 fails to properly manage GPU KV cache block ownership when concurrent child requests share a single transfer ID in prefill/decode disaggregated deployments. Attackers can trigger GPU memory exhaustion by submitting completion requests with multiple prompts, causing orphaned KV cache blocks to accumulate until process restart and eventually preventing legitimate requests from executing.
CVE-2026-94626
vLLM through 0.29.0 fails to validate the tp_size parameter in kv_transfer_params on OpenAI-compatible completion endpoints, allowing attackers to allocate unbounded memory. Attackers can supply arbitrary tp_size values in prefill/decode disaggregated deployments to exhaust memory and trigger kernel OOM-kill of the decode worker process.
CVE-2026-94625
vLLM through 0.29.0 contains a resource exhaustion vulnerability in MooncakeConnector where rejected prefill requests create ownerless transfer placeholders that are never reclaimed. Attackers can send rejected requests to exhaust sender task pools, causing valid requests to be delayed by up to 480 seconds while health checks continue returning success.