1  info:
 2    name: llama-cpp
 3    cve: CVE-2026-27940
 4    summary: llama.cpp integer overflow in mem_size calculation leads to heap buffer overflow and arbitrary code execution via malicious GGUF file
 5    details: >-
 6      llama.cpp is a C/C++ inference engine for LLM models. Prior to build b8146,
 7      gguf_init_from_file_impl() in gguf.cpp contains an integer overflow
 8      vulnerability in the mem_size calculation. While CVE-2025-53630 added an
 9      overflow check on per-addition accumulation of ctx->size, the final mem_size
10      computation at (n_tensors+1)*ggml_tensor_overhead() + ctx->size was not
11      guarded. When ctx->size is near SIZE_MAX (achievable via two large I8 tensors
12      whose per-addition sizes pass the CVE-2025-53630 check), the final addition
13      wraps around to a small value, resulting in an undersized heap allocation.
14      The subsequent fread() writes 528+ bytes of attacker-controlled data past the
15      buffer boundary. By tuning the allocation size to fall within glibc tcache
16      range, this escalates to full arbitrary code execution via system(/bin/sh).
17      Attack vector is local (AV:L); exploitation requires a user to load a
18      malicious GGUF model file. This is a bypass of the CVE-2025-53630 fix in
19      the same function. No public PoC; exploit maturity is LOW.
20    cvss: CVSS:3.1/AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H
21    severity: HIGH
22    security_advise: Upgrade llama.cpp to build b8146 or later. The fix adds an overflow guard to the mem_size calculation in gguf_init_from_file_impl() to prevent integer wrap-around when ctx->size is near SIZE_MAX. Avoid loading GGUF model files from untrusted sources.
23    references:
24    - https://github.com/ggml-org/llama.cpp/security/advisories/GHSA-3p4r-fq3f-q74v
25    - https://nvd.nist.gov/vuln/detail/CVE-2026-27940
26  rule: version < "b8146"
27  references:
28  - https://github.com/ggml-org/llama.cpp/security/advisories/GHSA-3p4r-fq3f-q74v
29  - https://nvd.nist.gov/vuln/detail/CVE-2026-27940