Data integrity
Data integrity is the property that data has not undergone unintended or unauthorized alteration, destruction or loss.
Scope: Integrity concerns preservation of data, not whether the original values were true. A faithfully stored incorrect measurement can remain unchanged. State the protected data, allowed changes and observation boundary. [1] [2]
Storage checks: A checksum over a block can match an old block after a lost write. Identity and version checks answer different questions, and detecting a mismatch does not establish successful recovery. These distinctions are explicit in a model of single-error RAID protection. [3]
Trust: A mechanism intended to detect accidental errors does not necessarily protect against deliberate modification. If an attacker can replace both data and an unkeyed digest, a matching result does not establish authenticity. [2]
Use: Durability describes preservation over time and specified failures; availability describes access. Those goals overlap with integrity but do not replace a check that the returned object has the expected identity and state. [4]
Research boundary: FAST 2026 distinguishes a direct-I/O userspace PI interface from experimental filesystem-managed PI; the latter is not established as generally available by the paper. [5]
Relations
- related
- Checksum, Silent data corruption, Data durability, Storage availability, Metadata, End-to-end data protection, Lost write, Message authentication code (MAC)
Citations
- NIST CSRC glossary: data integrity; definitions and source attribution.
- RFC 4949: data integrity, data integrity service and checksum entries (Informational glossary).
- Krioukov et al. Parity Lost and Parity Regained. FAST 2008, sections 2.3 and 4.3-4.6.
- Hesela: Why a valid checksum can still return the wrong data.
- Gupta et al. Advancing Data Integrity in Linux. FAST 2026, sections 5.2, 6.1 and 8.
Sources & caveats
- Terminology synthesis, not a storage certification. NIST lists context-dependent definitions; RFC 4949 is Informational and its terminology recommendations are the author's, not an IETF standard.
- The FAST 2008 examples concern the paper's stated model; no new device experiment or failure-rate measurement is reported.