{"id":"CVE-2021-41220","aliases":["GHSA-gpfh-jvf9-7wg5","BIT-tensorflow-2021-41220","PYSEC-2021-412","PYSEC-2021-629","PYSEC-2021-827"],"title":"Use after free / memory leak in `CollectiveReduceV2`","summary":"Use after free / memory leak in `CollectiveReduceV2`","severity":"high","cvss":7.8,"cvssVector":"CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H","vendor":"tensorflow","product":"tensorflow","ecosystem":"pip","affected":["tensorflow >= 2.6.0, < 2.6.1","tensorflow-cpu >= 2.6.0, < 2.6.1","tensorflow-gpu >= 2.6.0, < 2.6.1"],"patched":["tensorflow 2.6.1","tensorflow-cpu 2.6.1","tensorflow-gpu 2.6.1"],"published":"2021-11-10","updated":"2026-07-08","source":"OSV","sourceUrl":"https://osv.dev/vulnerability/GHSA-gpfh-jvf9-7wg5","references":[{"url":"https://github.com/tensorflow/tensorflow/security/advisories/GHSA-gpfh-jvf9-7wg5"},{"url":"https://nvd.nist.gov/vuln/detail/CVE-2021-41220"},{"url":"https://github.com/tensorflow/tensorflow/commit/ca38dab9d3ee66c5de06f11af9a4b1200da5ef75"},{"url":"https://github.com/pypa/advisory-database/tree/main/vulns/tensorflow-cpu/PYSEC-2021-629.yaml"},{"url":"https://github.com/pypa/advisory-database/tree/main/vulns/tensorflow-gpu/PYSEC-2021-827.yaml"},{"url":"https://github.com/pypa/advisory-database/tree/main/vulns/tensorflow/PYSEC-2021-412.yaml"},{"url":"https://github.com/tensorflow/tensorflow"}],"tags":["osv","pip"],"epss":0.00211,"epssPercentile":0.11641,"ingestedAt":"2026-07-08T18:25:49.753Z","slug":"CVE-2021-41220","body":"## Overview\n\n### Impact\nThe [async implementation](https://github.com/tensorflow/tensorflow/blob/8d72537c6abf5a44103b57b9c2e22c14f5f49698/tensorflow/core/kernels/collective_ops.cc#L604-L615) of `CollectiveReduceV2` suffers from a memory leak and a use after free:\n\n```python\nimport tensorflow as tf\n  \ntf.raw_ops.CollectiveReduceV2(\n  input=[],\n  group_size=[-10, -10, -10],\n  group_key=[-10, -10],\n  instance_key=[-10],\n  ordering_token=[],\n  merge_op='Mul',\n  final_op='Div')\n``` \n\nThis occurs due to the asynchronous computation and the fact that objects that have been `std::move()`d from are still accessed:\n\n```cc\nauto done_with_cleanup = [col_params, done = std::move(done)]() {\n  done();\n  col_params->Unref();\n};\nOP_REQUIRES_OK_ASYNC(c,\n                     FillCollectiveParams(col_params, REDUCTION_COLLECTIVE,\n                                          /*group_size*/ c->input(1),\n                                          /*group_key*/ c->input(2),\n                                          /*instance_key*/ c->input(3)),\n                     done);\n```\n\nHere, `done` is already moved from by the time `OP_REQUIRES_OK_ASYNC` macro needs to invoke it in case of errors. In this case, we get an undefined behavior, which can manifest via crashes, `std::bad_alloc` throws or just memory leaks.\n\n### Patches\nWe have patched the issue in GitHub commit [ca38dab9d3ee66c5de06f11af9a4b1200da5ef75](https://github.com/tensorflow/tensorflow/commit/ca38dab9d3ee66c5de06f11af9a4b1200da5ef75).\n\nThe fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, as this version is the only one that is also affected.\n\n### For more information\nPlease consult [our security guide](https://github.com/tensorflow/tensorflow/blob/master/SECURITY.md) for more information regarding the security model and how to contact us with issues and questions.\n\n### Attribution\nThis vulnerability has been reported by members of the Aivul Team from Qihoo 360.\n\n\n## Affected packages\n\n- `tensorflow >= 2.6.0, < 2.6.1`\n- `tensorflow-cpu >= 2.6.0, < 2.6.1`\n- `tensorflow-gpu >= 2.6.0, < 2.6.1`\n\n## Remediation\n\nUpgrade to a patched release:\n\n- `tensorflow 2.6.1`\n- `tensorflow-cpu 2.6.1`\n- `tensorflow-gpu 2.6.1`","depth":"twilight","depthScore":43,"depthScoreParts":{"impact":42.9,"likelihood":0,"exploitation":0,"ransomware":0},"changes":[]}