Get started →

OBSERVE · PRESERVE · CONTROL

Development

Progress, research, and qualification notes from the workbench.

Hexrook 2.0.0-dev28: managed-source integrity research

Dev28 completes the bounded Darwin content-integrity research milestone for managed original APFS, case-sensitive APFS and HFS+ sources. It adds a reusable acquisition session, original-property verification and a surviving owner for tested custodian failures. Ordinary macOS and Linux provider admission remain disabled; this is not a production macOS collector release.

A declared acquisition mechanism

The owner creates and retains protected backing before an ordinary producer receives its writable mounted subtree. A successful non-forced unmount ends the writable interval. The owner then binds a read-only view, retained content/property observations and consumer completion to that same backing. This preserves original-source identity instead of treating a prepared copy as proof of faithful acquisition. Arbitrary existing or previously exposed images are unsupported because later permission changes cannot establish their earlier write-custody history.

Content, names and original properties

Retained streams use the existing SHA-256 and BLAKE3 providers. Tests cover raw names, hard links, sparse and large content, timestamps, ACLs, extended attributes and resource forks. The new independent native source oracle compares original birth time, generation and every ACL with descriptor-based reader observations. Required property failures remain explicit failures. Allocation counts remain evidence-only. The retained-reader gate includes actual 4,294,967,313-byte streams; that is separate from the 256 MiB managed-session images.

Surviving tested custodian failures

One separately approved temporary Mac session ran fifteen volume lifecycles: three successful acquisitions and twelve deliberately failed acquisitions. Custodians were terminated at deterministic barriers before detach, after detach, after sealing and after publication. The surviving owner reaped each child and cleaned its verified resources; cleanup never converted a failed observation into success. The fixed coordinator accepted only run, status and revoke requests from the authorized peer, then removed its installation. It is not a persistent service or a way to bypass local authentication.

Native and retained-evidence validation

The ordinary native release, debug and sanitizer checkpoint exercised forty-five volume lifecycles and thirty-six custodian-loss injections. The separate privileged batch produced twelve supervised reader transcripts and ninety-six entry comparisons. Saved-session replay passed two positive and 734 negative cases on each of Windows, Linux and macOS. A reviewed saved-evidence defect was fixed: missing final verification or after-observation records can no longer pass after outer byte counts are repaired. These replay results check supplied evidence; they do not authenticate its host or repeat native acquisition on three platforms.

Compatibility checkpoint

Fresh Windows and experimental Linux release/debug reference matrices pass 198 and 145 summary entries respectively, with unchanged recorded inputs. These entries include builds and grouped tests, not that number of independent scans. Exact tested Windows binaries are promoted locally with previous copies retained. Database formats, existing filename compatibility and collector/core/storage boundaries are preserved. New databases continue to use the .hexrook extension. No public binary release or new production provider admission accompanies this development note.

Preserved findings and next work

The reproduced APFS snapshot omissions and clone/live-read counterexamples remain valid. Managed-source acquisition does not repair those mechanisms. Existing-source intake, death inside provider operations, simultaneous owner loss, restart recovery, arbitrary descendant containment and hard uninterruptible-I/O bounds remain unqualified. Production durable health and baseline integration, Linux L19 and Windows owner-death/deployment retain separate gates. Broader actual filesystem and mount compatibility remains a development goal.

Reproducibility and scope

The repository retains frozen sources, candidate hashes, dependency manifests, native command results, failed attempts, cleanup receipts and an explicit original-requirement acceptance matrix. Tests used pinned BlitzMax-NG 0.165.3.61 and the existing authorized Windows, Ubuntu and macOS environments. Native sanitizer coverage is limited to the recorded C drivers and worker paths, not the whole BlitzMax or Python runtime. No toolchain upgrade, persistent privilege installation, specialist scan or upstream submission was performed in this run.

Need more help?Find guidance, platform details, and troubleshooting.
HEXROOK | PRIVACY · CONTROL · FREEDOMINDEPENDENT TOOLS FOR A MORE SECURE WORLD.