Skip to content
Heist/ feed
Back to feed
Critical

vLLM Allows Remote Code Execution via PyNcclPipe Communication Service

Type: advisorySources: OSV

vLLM, an inference and serving engine for large language models (LLMs), has an issue in versions 0.6.5 through 0.8.4 that ONLY impacts environments using the `PyNcclPipe` KV cache transfer integration with the V0 engine. No other configurations are affected. vLLM supports the use of the `PyNcclPipe` class to establish a peer-to-peer communication domain for data transmission between distributed nodes. The GPU-side KV-Cache transmission is implemented through the `PyNcclCommunicator` class, while CPU-side control message passing is handled via the `send_obj` and `recv_obj` methods on the CPU side.​ The intention was that this interface should only be exposed to a private network using the IP address specified by the `--kv-ip` CLI parameter. The vLLM documentation covers how this must be limited to a secured network. The default and intentional behavior from PyTorch is that the `TCPStore` interface listens on ALL interfaces, regardless of what IP address is provided. The IP address given was only used as a client-side address to use. vLLM was fixed to use a workaround to force the `TCPStore` instance to bind its socket to a specified private interface. As of version 0.8.5, vLLM limits the `TCPStore` socket to the private interface as configured.

Products
Not specified
Affected versions
Not specified
Fixed versions
Not specified
CVEs
CVE-2025-47277
Confirmed fixed CVEs
Not specified
Updated in this feed
3 Nov 2026, 03:57 UTC · Revision 1

Source evidence

vLLM Allows Remote Code Execution via PyNcclPipe Communication Service

core-osv · Original source

Published 20 May 2025, 17:32 UTC

vLLM, an inference and serving engine for large language models (LLMs), has an issue in versions 0.6.5 through 0.8.4 that ONLY impacts environments using the `PyNcclPipe` KV cache transfer integration with the V0 engine. No other configurations are affected. vLLM supports the use of the `PyNcclPipe` class to establish a peer-to-peer communication domain for data transmission between distributed nodes. The GPU-side KV-Cache transmission is implemented through the `PyNcclCommunicator` class, while CPU-side control message passing is handled via the `send_obj` and `recv_obj` methods on the CPU side.​ The intention was that this interface should only be exposed to a private network using the IP address specified by the `--kv-ip` CLI parameter. The vLLM documentation covers how this must be limited to a secured network. The default and intentional behavior from PyTorch is that the `TCPStore` interface listens on ALL interfaces, regardless of what IP address is provided. The IP address given was only used as a client-side address to use. vLLM was fixed to use a workaround to force the `TCPStore` instance to bind its socket to a specified private interface. As of version 0.8.5, vLLM limits the `TCPStore` socket to the private interface as configured.

osv:CVE-2025-47277
Complete source record
{
  "id": "8e1269327c14867f6965ec07a5778f7f2a1da01abb2f95aaa0fc62e064e0dfa4",
  "url": "https://osv.dev/vulnerability/CVE-2025-47277",
  "method": "structured",
  "payload": {
    "kind": "advisory",
    "title": "vLLM Allows Remote Code Execution via PyNcclPipe Communication Service",
    "summary": "vLLM, an inference and serving engine for large language models (LLMs), has an issue in versions 0.6.5 through 0.8.4 that ONLY impacts environments using the `PyNcclPipe` KV cache transfer integration with the V0 engine. No other configurations are affected. vLLM supports the use of the `PyNcclPipe` class to establish a peer-to-peer communication domain for data transmission between distributed nodes. The GPU-side KV-Cache transmission is implemented through the `PyNcclCommunicator` class, while CPU-side control message passing is handled via the `send_obj` and `recv_obj` methods on the CPU side.​ The intention was that this interface should only be exposed to a private network using the IP address specified by the `--kv-ip` CLI parameter. The vLLM documentation covers how this must be limited to a secured network. The default and intentional behavior from PyTorch is that the `TCPStore` interface listens on ALL interfaces, regardless of what IP address is provided. The IP address given was only used as a client-side address to use. vLLM was fixed to use a workaround to force the `TCPStore` instance to bind its socket to a specified private interface. As of version 0.8.5, vLLM limits the `TCPStore` socket to the private interface as configured.",
    "evidence": "osv:CVE-2025-47277",
    "products": [],
    "severity": "critical",
    "structured": {
      "id": "CVE-2025-47277",
      "aliases": [
        "CVE-2025-47277",
        "GHSA-hjq4-87xh-g4fv",
        "PYSEC-2026-567"
      ],
      "metrics": [
        {
          "type": "CVSS_V3",
          "score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H"
        }
      ],
      "affected": [
        {
          "ranges": [
            {
              "repo": "https://github.com/vllm-project/vllm",
              "type": "GIT",
              "events": [
                {
                  "introduced": "2d1b9baa8f57fc59912c7bcd07fd630fb9d72c9d"
                },
                {
                  "fixed": "ba41cc90e8ef7f236347b2f1599eec2cbb9e1f0d"
                },
                {
                  "fixed": "0d6e187e88874c39cda7409cf673f9e6546893e7"
                }
              ],
              "database_specific": {
                "extracted_events": [
                  {
                    "cpe": "cpe:2.3:a:vllm:vllm:*:*:*:*:*:*:*:*",
                    "range": [
                      {
                        "introduced": "0.6.5"
                      },
                      {
                        "fixed": "0.8.5"
                      }
                    ],
                    "source": "CPE_RANGE"
                  },
                  {
                    "range": [
                      {
                        "fixed": "0d6e187e88874c39cda7409cf673f9e6546893e7"
                      }
                    ],
                    "source": "REFERENCES_COMMIT"
                  }
                ]
              }
            }
          ],
          "versions": [
            "v0.8.4",
            "v0.8.3rc1",
            "v0.8.2",
            "v0.8.1",
            "v0.8.0rc2",
            "v0.8.0rc1",
            "v0.7.3",
            "v0.7.2",
            "v0.7.1",
            "v0.7.0",
            "v0.6.6.post1",
            "v0.6.6",
            "v0.6.5"
          ],
          "database_specific": {
            "source": "https://storage.googleapis.com/cve-osv-conversion/osv-output/CVE-2025-47277.json"
          }
        }
      ],
      "modified": "2026-10-28T02:30:25.949609976Z",
      "provider": "osv",
      "references": [
        "https://docs.vllm.ai/en/latest/deployment/security.html",
        "https://github.com/CVEProject/cvelistV5/tree/main/cves/2025/47xxx/CVE-2025-47277.json",
        "https://github.com/vllm-project/vllm/security/advisories/GHSA-hjq4-87xh-g4fv",
        "https://nvd.nist.gov/vuln/detail/CVE-2025-47277",
        "https://github.com/vllm-project/vllm/commit/0d6e187e88874c39cda7409cf673f9e6546893e7",
        "https://github.com/vllm-project/vllm/pull/15988"
      ],
      "provider_metadata": {
        "cwe_ids": [
          "CWE-502"
        ],
        "cna_assigner": "GitHub_M",
        "osv_generated_from": "https://github.com/CVEProject/cvelistV5/tree/main/cves/2025/47xxx/CVE-2025-47277.json"
      }
    },
    "related_cves": [
      {
        "value": "CVE-2025-47277",
        "evidence": "/id"
      }
    ],
    "security_relevant": true,
    "severity_evidence": "/severity/0/score"
  },
  "source_id": "core-osv",
  "document_id": 492300,
  "published_at": "2025-05-20T17:32:27.034+00:00"
}