CVE-2025-52566

一、漏洞信息
 漏洞编号：[CVE-2025-52566](https://nvd.nist.gov/vuln/detail/CVE-2025-52566)
 漏洞归属组件：[llama.cpp](https://gitee.com/src-openeuler/llama.cpp)
 漏洞归属的版本：20230815,20240531,20241210
 CVSS V3.0分值：
  BaseScore：8.8 High
  Vector：CVSS：3.0/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H
 漏洞简述：
  llama.cpp is an inference of several LLM models in C/C++. Prior to version b5721, there is a signed vs. unsigned integer overflow in llama.cpp s tokenizer implementation (llama_vocab::tokenize) (src/llama-vocab.cpp:3036) resulting in unintended behavior in tokens copying size comparison. Allowing heap-overflowing llama.cpp inferencing engine with carefully manipulated text input during tokenization process. This issue has been patched in version b5721.
 漏洞公开时间：2025-06-24 12:15:46
 漏洞创建时间：2025-06-24 15:42:43
 漏洞详情参考链接：
  https://nvd.nist.gov/vuln/detail/CVE-2025-52566
<details>
<summary>更多参考(点击展开)</summary>

| 参考来源 | 参考链接 | 来源链接 |
| ------- | -------- | -------- |
|  | https://github.com/ggml-org/llama.cpp/commit/dd6e6d0b6a4bbe3ebfc931d1eb14db2f2b1d70af |  |
|  | https://github.com/ggml-org/llama.cpp/security/advisories/GHSA-7rxv-5jhh-j6xx |  |
|  | https://nvd.nist.gov/vuln/detail/CVE-2025-52566 |  |
|  | https://github.com/ggml-org/llama.cpp/commit/dd6e6d0b6a4bbe3ebfc931d1eb14db2f2b1d70af |  |
|  | https://www.mend.io/vulnerability-database/CVE-2025-52566 |  |
|  | https://access.redhat.com/security/cve/cve-2025-52566 |  |
|  | https://nvd.nist.gov/vuln/detail/CVE-2025-52566 |  |
|  | https://security-tracker.debian.org/tracker/CVE-2025-52566 |  |
|  | https://security-tracker.debian.org/tracker/CVE-2025-52566 |  |
|  | https://github.com/ggml-org/llama.cpp/commit/dd6e6d0b6a4bbe3ebfc931d1eb14db2f2b1d70af |  |
|  | https://github.com/ggml-org/llama.cpp/security/advisories/GHSA-7rxv-5jhh-j6xx |  |

</details>

漏洞分析指导链接：
  https://gitee.com/openeuler/cve-manager/blob/master/cve-vulner-manager/doc/md/manual.md
 漏洞数据来源:
  七彩瞬析开源风险感知平台
 漏洞补丁信息：
  <details>
<summary>详情(点击展开)</summary>

| 影响的包 | 修复版本 | 修复补丁 | 问题引入补丁 | 来源  |
| ------- | -------- | ------- | -------- | --------- |
| ggml-org/llama.cpp |  | https://github.com/ggml-org/llama.cpp/commit/dd6e6d0b6a4bbe3ebfc931d1eb14db2f2b1d70af.patch |  | ljqc |
|  |  | https://github.com/ggml-org/llama.cpp/commit/dd6e6d0b6a4bbe3ebfc931d1eb14db2f2b1d70af |  | nvd |
|  |  | https://github.com/ggml-org/llama.cpp/commit/dd6e6d0b6a4bbe3ebfc931d1eb14db2f2b1d70af |  | whitesource |
|  |  | https://github.com/ggml-org/llama.cpp/commit/dd6e6d0b6a4bbe3ebfc931d1eb14db2f2b1d70af |  | snyk |

</details>

二、漏洞分析结构反馈
 影响性分析说明：
  在版本 b5721 之前，llama.cpp 的词法分析器实现（llama_vocab::tokenize）（src/llama-vocab.cpp:3036）中存在一个有符号和无符号整数溢出，导致在标记复制大小比较中产生非预期行为。允许在词法分析过程中使用精心操纵的文本输入来使 llama.cpp 推理引擎堆溢出。此问题已在版本 b5721 中得到修复。
 openEuler评分：
  8.8
 Vector：CVSS：3.0/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H
 受影响版本排查(受影响/不受影响)：
  1.master(20241210):受影响
2.openEuler-24.03-LTS(20230815):受影响
3.openEuler-24.03-LTS-Next(20230815):受影响
4.openEuler-24.03-LTS-SP1(20230815):受影响
5.openEuler-24.03-LTS-SP2(20251009):受影响
6.openEuler-20.03-LTS-SP4:不受影响
7.openEuler-22.03-LTS-SP3:不受影响
8.openEuler-22.03-LTS-SP4:不受影响

修复是否涉及abi变化(是/否)：
  1.master(20241210):否
2.openEuler-20.03-LTS-SP4:否
3.openEuler-22.03-LTS-SP3:否
4.openEuler-22.03-LTS-SP4:否
5.openEuler-24.03-LTS(20230815):否
6.openEuler-24.03-LTS-Next(20230815):否
7.openEuler-24.03-LTS-SP1(20230815):否
8.openEuler-24.03-LTS-SP2(20251009):否

原因说明：
  1.master(20241210):正常修复
2.openEuler-24.03-LTS(20230815):正常修复
3.openEuler-24.03-LTS-Next(20230815):正常修复
4.openEuler-24.03-LTS-SP1(20230815):正常修复
5.openEuler-24.03-LTS-SP2(20251009):正常修复
6.openEuler-20.03-LTS-SP4:不受影响-漏洞代码不存在
7.openEuler-22.03-LTS-SP3:不受影响-漏洞代码不存在
8.openEuler-22.03-LTS-SP4:不受影响-漏洞代码不存在

三、漏洞修复
安全公告链接：https://www.openeuler.org/zh/security/safety-bulletin/detail/?id=openEuler-SA-2025-2373

src-openEuler/llama.cpp

内容风险标识

评论 (21)

src-openEuler/llama.cpp .gitee-modal { width: 500px !important; }

内容风险标识

CVE-2025-52566

评论 (21)

搜索帮助

src-openEuler/llama.cpp