About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.
If the files in a backup folder on a ReFS volume add up to more than the volume's capacity, the files are probably sharing physical space. ReFS lets several files point at the same clusters through block cloning, so the sum of file sizes (logical size) can be much larger than the space used. The disk is not lying. But a copy to another volume cannot share clusters, so it needs the full logical size.
What the symptom means
You see a folder whose files total, say, 28 TB on a volume of 8 TB, with free space that is still falling slowly. Or a copy of the folder to a bigger disk fails with "not enough space" even though the source volume is not full.
Microsoft's block cloning page explains the mechanism. Block cloning "instructs the file system to copy a range of file bytes on behalf of an application". On ReFS the copy is "a low-cost metadata operation rather than reading from and writing to file data", because "ReFS enables multiple files to share the same logical clusters (physical locations on a volume)". It adds that "identical data isn't physically stored multiple times, improving storage capacity".
To track sharing, ReFS "introduces reference counts into its metadata for copied regions", recording "the number of distinct file regions that reference the same physical regions". A new write to a shared region is not made in place. The page says "writes to shared regions trigger an allocate-on-write mechanism, where ReFS allocates a new region for the incoming write". That keeps every file's content separate, and it is why one file can change without changing another.
Block cloning is available on ReFS from Windows Server 2016. Applications call it through FSCTL_DUPLICATE_EXTENTS_TO_FILE. ReFS also lists backup targets as a supported use: "applications and hardware that implements its own resiliency and availability solutions" can use "integrity-streams, block-cloning". Backup software that builds a new full restore point by cloning unchanged data from the previous one will produce exactly this symptom.
Likely causes
| Cause | What you see | How to confirm |
|---|---|---|
| Backup software clones unchanged data into each new restore point | Many large files; the sum of sizes far above used space | Sum of file lengths is above Size minus SizeRemaining for the volume |
| Windows copies on the same volume are cloned natively | A manual copy of a big folder costs almost no space | Free space barely moves after a same-volume copy; builds with native clone support (see Risks) |
| ReFS deduplication (Azure Local) is enabled | Savings listed on the volume; used space below logical size | Get-ReFSDedupStatus -Volume <path> shows Enabled : True, Type : Dedup |
| Windows Data Deduplication is enabled on the ReFS volume | Optimised files; SavingsRate reported |
Get-DedupStatus -Volume <drive> returns savings fields |
All four share one effect. Some bytes are counted in more than one file but stored once.
Prove the cause before changing anything
Every check below only reads. Do the first three before anything else.
1. Real use of the volume. Get-Volume returns Size and SizeRemaining for each volume (Microsoft's example shows a volume with SizeRemaining 23.61 GB and Size 465.42 GB). Used space is Size minus SizeRemaining.
Get-Volume -DriveLetter D
2. The logical total of the folder. Measure-Object sums a property. Microsoft's example measures file sizes with -Property Length -Sum.
Get-ChildItem -Path D:\Backups -Recurse -File | Measure-Object -Property Length -Sum
3. The cluster size of the ReFS volume. The Win32 block cloning page gives this command and says the cluster size is fixed at format time:
fsutil fsinfo refsinfo D:
Compare the numbers. If the logical total from step 2 is larger than used space from step 1, the difference is data counted in several files but stored once. A good result is that the volume's real use is well below its Size. A bad result is real use close to Size: a volume that is truly nearly full, whatever the files add up to.
4. Which feature is in use. On Azure Local, Get-ReFSDedupStatus -Volume <path> | FL shows whether ReFS deduplication is on. On Windows Server with Data Deduplication, Get-DedupStatus -Volume D: returns status for volumes with deduplication metadata. These pages list the commands; the output depends on your system, so read the fields rather than expecting a layout.
5. The backup product's own setting. Ask the vendor whether it uses block cloning for synthetic fulls, merges or "fast" restore points on ReFS. Microsoft's pages do not describe any backup product's behaviour.
Worked numbers
Example (illustrative): backup01 has a ReFS volume D: with Size of 8 TB (decimal). A backup product writes one full restore point of 4 TB, then six more that each look like a 4 TB full file but clone the unchanged data from the previous one and add 0.5 TB of new data.
- Logical total: 7 × 4 = 28 TB. This is what the sum of file lengths shows.
- Real use: 4 + 6 × 0.5 = 7 TB.
SizeRemainingshows 8 − 7 = 1 TB free. - Shared data: 28 − 7 = 21 TB, counted in files but stored once.
- A copy to another volume: writes 28 TB. A target of 20 TB fails, and a target of 30 TB succeeds. The same 7 TB of unique data has grown fourfold (28 ÷ 7 = 4).
The folder looks 3.5 times larger than the volume (28 ÷ 8 = 3.5), but the volume is 7 ÷ 8 = 87.5% full, so free space is the true limit. The 1 TB left is the figure to alert on.
Fix
The sharing itself is rarely a fault, so most of the work is planning around it.
Cause 1 and 2: cloned data from backup software or Windows
- Watch
SizeRemaining, not the sum of file lengths. Treat file totals as a logical figure only. - Size any copy target at the logical total. The block cloning page requires that "the source and destination files must be on the same ReFS volume", so a copy to another volume cannot share clusters. It follows that the copy writes every byte. In the worked numbers above, 28 TB logical needs 28 TB of space at the destination, not the 7 TB the source volume uses. This is a consequence of the documented restriction, not a statement printed on the page.
- Use the backup software's own export, copy or archive function when moving restore points, and ask whether it preserves sharing. Microsoft does not say.
Cause 3: ReFS deduplication
Disruptive step: turning ReFS deduplication off does not recover space and does not revert it at once. The Azure Local page says disabling "doesn't undo deduplication, as all the operations occur at the metadata layer. Over time, deduplicated data returns to its original state as the volume incurs reads and writes." Used space then rises as shared data is rewritten, so do not disable it on a volume that has no room for the full logical size. To back out, Enable-ReFSDedup -Volume <path> -Type Dedup turns it on again, but data already rewritten will need a new job to be shared again.
Cause 4: Windows Data Deduplication on ReFS
Disruptive step: Microsoft's way to turn off Data Deduplication and undo volume optimisation is the Unoptimization job:
Start-DedupJob -Type Unoptimization -Volume <volume>
Microsoft says it "will fail if the volume does not have sufficient space to hold the unoptimized data". Check Size minus SizeRemaining against the logical total first. Microsoft documents no undo; to deduplicate again you would enable the feature and let new optimisation jobs run.
Risk: Microsoft states that "running Robocopy with Data Deduplication isn't recommended because certain Robocopy commands can corrupt the Chunk Store". Do not move a deduplicated volume's data with Robocopy. Windows Server Backup can back up an optimised volume "as-is".
Confirm the fix
- Re-run steps 1 and 2 above and write down used space and the logical total together, with the date.
- Before any copy, check that the destination volume has
SizeRemainingat least equal to the logical total, plus headroom. - After a copy, check that used space on the destination equals the logical total you planned for, and that the source
SizeRemaininghas not changed. - For ReFS deduplication,
Get-ReFSDedupStatusshould show the sameEnabledandTypeas before you started.
What to watch afterwards
- Free space on the volume, not the folder size. Alert when
SizeRemainingfalls below a set share ofSize. - Writes into shared regions. After cloning, a write into a shared cluster makes a new one. The Win32 block cloning page says "A write of any size to a shared cluster (even 1 byte) causes the entire cluster to be duplicated". So a write to a shared cluster copies the whole cluster, whatever the write size and whatever the cluster size is. A backup that keeps rewriting the middle of cloned files will use more space than the plan.
- Cluster size. ReFS offers 4 KB and 64 KB clusters. The overview recommends 4 KB for most deployments and 64 KB for large, sequential workloads. It cannot be changed without reformatting.
- Growth against the logical total. If used space rises toward the logical total, less is being shared than before.
Read next
Related articles you may want next.
- Backup & RecoveryBackup storage is bigger than the data it protects: where the extra space goesWhy backup storage can exceed the server it protects: retention, changed data, compression, unreclaimed space and chunk rounding, with arithmetic.Read article
- Backup & RecoveryDeduplication ratio explained: what 10:1 on a backup report means for the billWhat a 10:1 deduplication ratio on a backup report means, why it climbs with retention, and how to turn it into GB stored per GB protected.Read article
- Backup & RecoveryBacking up and restoring Data Deduplication volumes: what is really on the diskPlan Windows Data Deduplication recovery around the chunk store, supported backup modes, compatible servers and realistic restore capacity.Read article
Risks and limits
- These commands have not been run or lab-tested. They are taken from Microsoft documentation.
- Microsoft's pages do not say how
Get-ChildItem, File Explorer ordirreport a file that shares clusters, or how a backup product reports it. The logical total here is the file length, and used space isSizeminusSizeRemaining. - The block cloning page states that block cloning occurs natively in supported Windows copy operations "starting with Windows 11 24H2 and Windows Server 2025 builds". The Win32 page says "Starting with the Windows 11 Moment 5 update (KB5034848) and later releases of Windows client and Windows Server builds". The two pages name different builds, so check your own build before relying on native cloning.
- The page for
fsutil fsinfodoes not listrefsinfo. The Win32 block cloning page does, for cluster size only. Do not expect it to report shared space. - ReFS deduplication is described on an Azure Local page. Windows Data Deduplication on ReFS is supported "starting with Windows Server 2019". Microsoft says the two cannot run together.
- A copy needing the full logical size follows from the same-volume rule. Some products have their own copy or export tools that behave differently, and Microsoft's pages say nothing about them.
Sources
- Block cloning on ReFS: Microsoft Learn, page date 2024-09-23, updated 2026-02-16, accessed 17 November 2026
- Block Cloning (Win32 apps): Microsoft Learn, page date 2025-10-14, accessed 17 November 2026
- FSCTL_DUPLICATE_EXTENTS_TO_FILE: Microsoft Learn, page date 2018-12-05, accessed 17 November 2026
- Resilient File System (ReFS) overview: Microsoft Learn, page date 2025-07-22, accessed 17 November 2026
- Optimize storage with ReFS deduplication in Azure Local: Microsoft Learn, page date 2026-05-14, accessed 17 November 2026
- Data Deduplication interoperability in Windows Server: Microsoft Learn, page date 2025-06-30, accessed 17 November 2026
- Running Data Deduplication: Microsoft Learn, page date 2022-02-18, updated 2025-09-15, accessed 17 November 2026
- Get-DedupStatus: Microsoft Learn, Windows Server 2025 PowerShell reference, accessed 17 November 2026
- Get-Volume: Microsoft Learn, Windows Server 2025 PowerShell reference, accessed 17 November 2026
- Measure-Object: Microsoft Learn, PowerShell 7.6 reference, accessed 17 November 2026
- Get-ChildItem: Microsoft Learn, PowerShell 7.6 reference, accessed 17 November 2026
- Fsutil fsinfo: Microsoft Learn, page date 2017-10-16, updated 2026-02-16, accessed 17 November 2026
This article was originally published on .