CVE-2026-106547: 
HDF5 vulnerability analysis and mitigation

CVE-2026-106547 is a heap-based buffer overflow vulnerability in the H5VM_array_fill() function located in src/H5VM.c of The HDF Group's HDF5 library. It affects HDF5 versions from 1.10.0 up to (but not including) 2.2.0, and was published on October 6, 2026. A remote attacker can trigger an application crash or potentially execute arbitrary code by supplying a specially crafted HDF5 file. It carries a CVSS v4.0 base score of 8.5 (High) (GitHub Advisory, Feedly).

The root cause is a heap-based buffer overflow (CWE-122) in H5VM_array_fill(), triggered during the fill-value initialization process in H5D__fill_init(). When a dataset's unallocated chunks are read, the library fills a fill-value buffer using datatype and dataspace metadata embedded in the HDF5 file. If an attacker crafts the file so that this metadata is inconsistent with the buffer's allocated size, the write operation exceeds the buffer boundary on the heap. Critically, the attacker controls the content written past the buffer boundary through the fill value stored in the file, enabling potential code execution (GitHub Advisory, HDF5 PR #6529).

Successful exploitation can result in an application crash (denial of service) or arbitrary code execution in the context of the process parsing the malicious HDF5 file, with high impact to confidentiality, integrity, and availability of the vulnerable system. Because HDF5 is widely used in scientific computing, data analysis, and machine learning pipelines, any service or application that automatically ingests HDF5 files from untrusted sources is at elevated risk. Lateral movement potential depends on the privileges of the affected process, but code execution could allow an attacker to pivot within the host environment (GitHub Advisory, Feedly).

As of the time of disclosure, there is no public proof-of-concept exploit and no evidence of active in-the-wild exploitation (Feedly). The EPSS score is 0.0, indicating a currently low probability of exploitation in the near term. The vulnerability is not listed in the CISA Known Exploited Vulnerabilities (KEV) catalog. Exploitation requires user interaction (passive) — a victim application must open a crafted HDF5 file — but no privileges are required from the attacker's side (GitHub Advisory).

1. Craft a malicious HDF5 file: Create an HDF5 file containing a chunked dataset with a variable-length or other datatype. Embed inconsistent datatype/dataspace metadata such that the fill-value buffer size calculated by H5D__fill_init() is smaller than the data H5VM_array_fill() will attempt to write.
2. Embed a controlled fill value: Set the fill value in the file to attacker-controlled content. This content will be written past the end of the heap buffer, enabling heap corruption with controlled data.
3. Deliver the file to the target: Distribute the crafted HDF5 file to a target application or user — for example, via email attachment, file upload to a data processing service, or a shared network location where the application automatically ingests files.
4. Trigger file parsing: Cause the target application to open and read the malicious HDF5 file. When the application reads the unallocated chunk, H5D__fill_init() and H5VM_array_fill() are invoked, triggering the heap overflow.
5. Achieve crash or code execution: Depending on heap layout and memory protections, the overflow may crash the application (DoS) or, with further heap grooming, enable arbitrary code execution in the context of the parsing process (GitHub Advisory, HDF5 PR #6529).

• File System: Presence of unexpected or externally sourced .h5 or .hdf5 files in application input directories, especially those with unusual metadata structures or anomalous file sizes.
• Process: Application processes (e.g., Python, MATLAB, scientific data pipelines) crashing unexpectedly with segmentation faults or heap corruption errors when processing HDF5 files.
• Logs: Application crash logs or core dumps referencing H5VM_array_fill, H5D__fill_init, or H5VM.c in stack traces; error messages related to buffer overflows or memory access violations in HDF5 library output.
• Network: Unexpected inbound file transfers of .h5/.hdf5 files from external or untrusted sources to systems running HDF5-dependent applications.

Upgrade HDF5 to version 2.2.0 or later, which contains the fix introduced via pull request #6529 that reworks the nonexistent chunk read pathway to avoid the unsafe fill-value buffer write (HDF5 PR #6529, GitHub Advisory). As a workaround where upgrading is not immediately possible, restrict processing of HDF5 files to trusted sources only, and implement strict validation of HDF5 file metadata before passing files to the library. Consider sandboxing or isolating processes that parse HDF5 files from untrusted inputs to limit the blast radius of potential exploitation.