Measured tool behavior / synthetic byte patterns

QCOW2 dependency and verification observations

Ten controlled cases separate overlay identity, backing dependencies, image consistency and guest-byte comparison.

Metadata / Raw commands and hashes / Python harness / Analysis and interpretation

Method

Nine offline QCOW2 scenarios use four 65536-byte clusters and a known full-byte oracle. A tenth raw-image control compares unequal virtual sizes with a zero tail. 3 repetitions on one host; explicit formats, closed writers, no live VM. Every successful raw export is checked byte for byte against a separately constructed expectation. Outcomes matched across all three repetitions. These repeats are not a sample of operational failures.

How to interpret the result

Base replacement changed 131,072 guest-visible bytes while the overlay hash and consistency check still passed. The explicit-zero cluster stayed zero. Shortening the base to 128 KiB also passed check, but the inherited tail became zeros: an existing short base is different from an absent base. A separate one-byte payload change also passed check. Missing dependencies instead prevented opening; the dataset does not describe those cases as silent corruption.

For these commands, check exit 1 is an opening failure. Compare exit 1 means a completed difference result, while compare exit 2 is an opening failure. Do not merge these into a single generic failure label. Raw JSON retains standard output, standard error, tool versions and executable hashes.

Reproduce

Measured package: qemu-img version 8.2.2 (Debian 1:8.2.2+ds-0ubuntu1.18); Python 3.12.3; Linux 6.8.0-146-generic. The tools were extracted from the Ubuntu package without installing a service or running a VM. Use the named package for closest reproduction; the harness records other supplied executable versions rather than claiming they are equivalent.

python3 qcow2-dependencies.py --repetitions 3 > observations.json
python3 build-qcow2.py --input observations.json \
  --harness qcow2-dependencies.py --site replay --surface dev

Download the harness and generator. Both use the Python standard library; the experiment additionally needs qemu-img and qemu-io. Optional executable-path arguments and --temp-dir are supported. Only private temporary fixtures are manipulated. Each subprocess has a timeout. Generated CSV represents null values as empty cells; JSON preserves null and boolean types.

The generator rejects missing or duplicated scenarios, changed harness identity, unexpected exit codes, summary/command-receipt disagreement and incorrect payload hashes. It preserves the observation timestamp rather than substituting build time. The raw file and metadata carry SHA-256 provenance for the actual downloaded serialization. These checks detect inconsistencies; they are not authentication of an untrusted log. Upstream source review was pinned to QEMU v8.2.2; the tested binary additionally includes Ubuntu packaging patches.

Limits

• Synthetic bytes and controlled file mutations, not a VM restore, power-loss test, performance benchmark or failure-rate sample.
• One Ubuntu QEMU 8.2.2 package; other versions, options and formats can differ.
• Uncompressed, unencrypted images with standard clusters and one raw base; no internal snapshots, external data files or dirty bitmaps.
• Successful guest-byte export does not establish guest filesystem, application or multi-disk consistency.

Column definitions

repetition

Repeat number on one host, not an independent reliability trial.

case

Controlled fixture manipulation.

checkExit

qemu-img check status: 0 completed consistently; 1 could not open the missing dependency. Null for raw size control.

compareExit

qemu-img compare: 0 equal under default rules; 1 differs; 2 opening error.

exportExit

Full raw export: 0 success; 1 opening failure; null for size control.

topHashUnchanged

Overlay SHA-256 equals its own pre-manipulation hash; null for size control.

guestMatchesBaseline

Every exported byte equals the independent baseline; null when unreadable or not applicable.

changedGuestBytes

Differing byte positions against the equal-length baseline; null when unavailable.

strictCompareExit

Strict compare status in the zero-tail control; null elsewhere.

Sources and reuse

CC0-1.0 for original Hesela code and observations; QEMU and upstream documentation retain their own licenses.

1. QEMU 8.2.2: qemu-img check, compare and rebase
2. QEMU 8.2.2: QCOW2 image file format, cluster mapping and zero clusters
3. QEMU 8.2.2: qcow2_check_refcounts implementation
4. libvirt: disk image chains, backing formats and relocation

Related definitions: QCOW2, QCOW2 backing chain, copy-on-write and backup verification.