Skip to content

Releases

ntfs2refs 0.0.1

Choose a tag to compare

@github-actions github-actions released this 18 Nov 19:08
87517a1

ntfs2refs 0.0.1

WARNING: NTFS2REFS IS EXPERIMENTAL SOFTWARE.
FOR EVALUATION AND TESTING ONLY. DO NOT USE IT IN PRODUCTION.

IT REWRITES THE FILE SYSTEM OF A VOLUME IN PLACE, FROM NTFS TO REFS.
IT HAS BEEN TESTED ONLY ON VIRTUAL DISKS IN A LABORATORY. A FAULT IN
IT, IN WINDOWS OR IN THE HARDWARE CAN DESTROY EVERYTHING ON THE VOLUME.

BACK UP EVERYTHING ON THE VOLUME, AND CHECK THAT YOU CAN RESTORE THE
BACKUP, BEFORE YOU CONVERT. YOU USE THIS SOFTWARE ENTIRELY AT YOUR OWN
RISK. IT COMES WITH NO WARRANTY OF ANY KIND. READ DISCLAIMER.md.

Built by lab\New-Release.ps1 from commit 87517a1cdf90ba5a3f78765548406a08aab9a340 with rustc 1.98.1 (48a229cea 2026-09-01).

File SHA-256 Signed by
ntfs2refs.exe 0307eb8c902689ff06437975adcccf9f945a40645905d89f1977f942f4e5b4d6 Fara Technologies LLC
ntfs2refs-recovery.exe 07b512407fb284a1653ab1f6348acbde713089542fb7f8e97fa31e5c040d9738 Fara Technologies LLC
lab-volume-manifest.exe afd9afa48885518564f8195dabd5a00f0b67d49f6f1e4bb649ab5ab895f06ef2 Fara Technologies LLC

What this executable admits

capabilities.json in this archive is the output of ntfs2refs capabilities (schema 5). In short:

  • Windows build 26100 or later; conversions were run on builds 26100 and 26300.
  • ReFS clusters of 4096 or 65536 bytes for a data volume; the version Windows formats the shadow with: 3.14 or a later 3.x (only 3.14 has been converted).
  • A data volume: a disk other than the running Windows': a GPT disk, which may hold other partitions, or an MBR disk the volume has to itself; --list says for each NTFS partition whether it qualifies.
  • An installed Windows that is not running: windows-11-build-26300-dev-drive.

Not carried over:

  • object IDs of files and directories, which Windows gives to the targets of shortcuts: ReFS has no object IDs.
  • the NTFS compression flag: compressed data is stored as ordinary data on ReFS.
  • the change journal: the converted volume starts a new one.
  • the case-sensitive mark of a directory in which no two names differ only by case: the converter does not set it on the new volume; every name stays reachable.

Refused before anything is written:

  • 8.3 short names. To do: fsutil 8dot3name strip, then fsutil 8dot3name set : 1.
  • a BitLocker-encrypted volume. To do: decrypt it (manage-bde -off).
  • shadow copies of the volume, snapshot storage kept on it, or a shadow-copy provider of backup software that holds a shadow copy on the machine or does not answer. To do: delete the shadow copies; the provider is named.
  • EFS-encrypted files. To do: cipher /d.
  • files with the TEMPORARY or OFFLINE attribute.
  • reparse points other than symbolic links, junctions, app execution aliases and the files of the Windows Subsystem for Linux (Data Deduplication, compact /exe, cloud placeholders). To do: named per kind.
  • a named data stream on a directory.
  • disk quotas in use. To do: fsutil quota disable.
  • two names in one directory that differ only by case. To do: rename one of them, then fsutil file setCaseSensitiveInfo disable.
  • a dirty volume, an NTFS log that was not closed, an open transaction. To do: chkdsk /f, or a clean dismount.
  • a volume with no file on it. To do: format it as ReFS.

What was shown, and what was not

docs/feature-matrix.md lists every NTFS feature with what a conversion does with it and the lab case that shows it; docs/recovery-matrix.md lists every point at which a conversion was interrupted and how each recovery ended. docs/cli-workflow.md is the operating and recovery guide.

Not shown:

  • a physical disk: every lab disk is virtual.
  • a volume holding more than 23 GB.
  • a power cut to the storage itself: the lab ends the converter's process or turns the virtual machine off.
  • a Windows build other than those in windows_builds.confirmed_builds, and a ReFS version other than 3.14.
  • a conversion to standard ReFS on Windows 11 (the lab's does not format one), a Windows boot volume of Windows Server, and Windows 11 on build 26100.
  • --minimize-scratch.
  • the restart route (--restart-media) on Windows 11, on a physical machine, with BitLocker on the Windows volume, and interrupted while it converts.

Verify the download

The executables are signed by Fara Technologies LLC and by Mike Fara (Azure Artifact Signing), and the ZIP and executables carry a build-provenance attestation of this repository's release workflow:

gh attestation verify ntfs2refs-0.0.1-windows-x64.zip --repo faratech/ntfs2refs --signer-workflow faratech/ntfs2refs/.github/workflows/release.yml

ntfs2refs 0.0.1-alpha: Windows boot volume, NTFS to ReFS in place (lab-qualified)

Choose a tag to compare

@faratech faratech released this 11 Nov 16:49

ntfs2refs 0.0.1-alpha

A private experimental pre-release. It is not a production release.

This build converts an installed Windows 11 system partition from NTFS to ReFS
in place, and the converted installation boots to the desktop. It is qualified
on one lab virtual machine only. Make an independent backup before using it.

Build

  • Source commit 3aaaf11bb4e5a4891942feeabc974d4188916e98 (branch
    codex/boot-volume), Rust 1.98.1, x86_64-pc-windows-msvc, release profile.
  • ntfs2refs.exe reports version 0.1.0. SHA-256
    e057c9e69554977167f47665c2a56425d0b77ab6f3a317f9174ed348af214492.
    It is the executable that ran qualification run full-16 below; the only
    commit after its build changed project documentation.
  • lab-volume-manifest.exe, the separate tool that lists a volume before and
    after conversion and compares the two listings. SHA-256
    61b509b7b71aa9fc9d99cb15464ac1a3c0ceef2af8e89992037b78562b3bce92.

What was verified

Lab machine: Hyper-V generation 2 VM, Secure Boot on, 8 virtual processors,
24 GB of memory. Source: Windows 11 build 26300 on disk 0 partition 3 (64 GiB
disk, 22.9 GB used, 143,580 objects, 112,609 8.3 short names).

Each full run starts from the NTFS baseline checkpoint, converts from a
separate control OS, compares complete before/after listings made by
lab-volume-manifest, then starts the VM with only disk 0 attached and signs
in at the console.

Run Build Conversion Listing Boot
full-16 this executable 28.0 min identical, 0 differences desktop; ReFS 3.14, 4 KiB clusters; winload.efi and refs.sys signatures valid
full-12 same source, earlier build 27.7 min identical desktop

The same source also passed the 1-in-40 fixture twice (identical listings) and
--recover of a run stopped after apply. cargo test --locked --workspace --all-targets at the release commit: 713 passed, 0 failed, 52 ignored.

Differences the listing comparison allows, with their counts on the lab
volume: 112,609 short names that became links; NTFS compression flag removed
(218 files); System Volume Information content (8); kernel cache attributes
dropped (17,140).

How it was run

From a separate Windows installation (the control OS), with the partition to
convert attached as a data volume and not in use. In the lab it runs as
SYSTEM:

ntfs2refs.exe --drive D --windows-boot-volume --scratch-directory S:\runs --target-format dev-drive

D: is the NTFS Windows partition to convert. S: is a separate volume for
working files; the full run used about 30 GB of it. The converter takes the
source offline and holds it read-only until it has prepared and checked
everything. Nothing is written to the source before that point; a run that
stops earlier leaves the source unchanged, and is started again.

Before its first write the converter prints a recovery command with a
checksum. Keep it. If the run stops after it has started writing, complete
the conversion with:

ntfs2refs.exe --recover <printed directory> --recovery-sha256 <printed checksum>

Known limits

  • Qualified on one VM and one Windows build. Not qualified on physical
    hardware, other Windows builds, BitLocker, other partition layouts or larger
    volumes.
  • Each 8.3 short name becomes a relative symbolic link beside its long name.
    These are visible in Explorer: PROGRA~1 in C:\, a MICROS~1 icon on the
    desktop.
  • The NTFS compression flag is not carried over; the files are stored
    uncompressed.
  • A power cut during preparation is not resumed: start again.
  • A full conversion of the lab volume takes 28 to 30 minutes on the lab host.

Assets

  • ntfs2refs-0.0.1-alpha-windows-x64.zip: ntfs2refs.exe, lab-volume-manifest.exe and these notes. SHA-256 4966456f4c0bac93095cd6112a0eb5f06262842b3965ae6c7e86845cf17d713e (also in the .sha256 file).
  • ntfs2refs-0.0.1-alpha-source.zip: git archive of tag 0.0.1-alpha. SHA-256 25bca3be312f3235334c5803cc6453236cddaa31cc8f537f02b33112409d72ca.

Recovered qualification records

The original full-16 console output and workspace test log have been recovered
and preserved in ntfs2refs-0.0.1-alpha-qualification.zip (SHA-256
6d7819d61cb0bfaa5b0cffbc73c137d876f31cc83bde811c3bc0eca29b9ab720).
The bundle contains source-line provenance and a hash manifest. The original
source and binary release assets above remain unchanged.

The original complete before/after JSONL files, verdict.json, raw trust-patch
records and desktop screenshot were not recovered. The bundle preserves actual
original command output; it does not fabricate those missing files. The separate
rec-04 run completed post-apply recovery and mounted healthy ReFS, but had no
before-listing and reported converted-not-compared. That result does not
establish full-content recovery, arbitrary interrupted writes or a Windows
boot-volume rollback matrix.

Integration closeout

The converter and corrected operating documentation were merged into main in
PR #27, commit
ab015130a04438a45df59b7601fbe0163dd84fb7.
Windows CI passed
formatting, Clippy, all workspace tests (713 passed, 52 ignored), and the complete
release build. The merged files exactly match the CI-tested merge tree.

Use the current operating guide
and qualification details.
The original release binaries, source archive and tag remain unchanged.

Additional evidence assets:

  • ntfs2refs-0.0.1-alpha-closeout-validation.zip: local checks and the independent
    1-in-10 fixture listings and comparison, with zero unexpected differences.
    SHA-256 6c5f41619236d4759b1b7cd99d1ea54ba49787ca9caad37e6db12cf067fdbf27.
  • ntfs2refs-0.0.1-alpha-ci-closeout.zip: complete hosted CI log, local test logs,
    the test-only profiler synchronization fix and merge-tree verification.
    SHA-256 c2590b486e04ef7d8fc4e970304d1fa1251d69d67e7cc26f356c4a9ac942448d.

These checks do not expand the native hardware or recovery qualification scope
of the frozen alpha.

Experimental native-shadow alpha (7feb4c0)

Choose a tag to compare

@faratech faratech released this 03 Nov 03:25

This is a private experimental prerelease for a narrowly qualified Windows data-volume profile. It is not a stable production release.

The exact executable is built from 7feb4c06d21e9de73bb88961197f5b745463e490 and reports version 0.1.0. Its SHA-256 is c7b5809f4456f55e71f19b47d4e4cb9b7ee6cad86c4ff0fac36c1db5a5b909d1.

The recorded Windows 11 VM case converted an existing 52 GiB NTFS partition containing 1,534 user objects; all 811 independent stream hashes matched. A subsequent mutation workload and clean reboot matched all 811 original streams, 1,534 namespace/type/length objects, 258 mutation files and 256 directory entries. Controlled first-write interruption, resume and rollback also passed for the recorded format-4 journal case.

The package includes operating and recovery instructions, explicit compatibility limits and evidence. It requires an independent backup and working storage; keep the printed recovery checksum. The observed profile used a dedicated nonboot GPT data disk and a native ReFS Dev Drive. This build does not qualify arbitrary hardware, boot volumes, standard ReFS formatting, all NTFS feature combinations or physical power loss.

Windows package SHA-256: dd9f2dcd169ff20a1701eb4ddf2cbbf9d453944a48d6969e35500a49cd26e0b3.
Source archive SHA-256: 58f90de936a09a76c00d31993d3803aa71f4b5a6a6aebbf0de809ab87ba783e9.